One document matched: draft-ietf-dccp-tfrc-voip-07.ps

%!PS-Adobe-3.0
%%Creator: groff version 1.18.1
%%CreationDate: Mon Nov 20 14:24:57 2006
%%DocumentNeededResources: font Courier-Bold
%%+ font Times-Bold
%%+ font Times-Roman
%%+ font Times-Italic
%%+ font Courier
%%DocumentSuppliedResources: procset grops 1.18 1
%%Pages: 39
%%PageOrder: Ascend
%%Orientation: Portrait
%%EndComments
%%BeginProlog
%%BeginResource: procset grops 1.18 1
/setpacking where{
pop
currentpacking
true setpacking
}if
/grops 120 dict dup begin
/SC 32 def
/A/show load def
/B{0 SC 3 -1 roll widthshow}bind def
/C{0 exch ashow}bind def
/D{0 exch 0 SC 5 2 roll awidthshow}bind def
/E{0 rmoveto show}bind def
/F{0 rmoveto 0 SC 3 -1 roll widthshow}bind def
/G{0 rmoveto 0 exch ashow}bind def
/H{0 rmoveto 0 exch 0 SC 5 2 roll awidthshow}bind def
/I{0 exch rmoveto show}bind def
/J{0 exch rmoveto 0 SC 3 -1 roll widthshow}bind def
/K{0 exch rmoveto 0 exch ashow}bind def
/L{0 exch rmoveto 0 exch 0 SC 5 2 roll awidthshow}bind def
/M{rmoveto show}bind def
/N{rmoveto 0 SC 3 -1 roll widthshow}bind def
/O{rmoveto 0 exch ashow}bind def
/P{rmoveto 0 exch 0 SC 5 2 roll awidthshow}bind def
/Q{moveto show}bind def
/R{moveto 0 SC 3 -1 roll widthshow}bind def
/S{moveto 0 exch ashow}bind def
/T{moveto 0 exch 0 SC 5 2 roll awidthshow}bind def
/SF{
findfont exch
[exch dup 0 exch 0 exch neg 0 0]makefont
dup setfont
[exch/setfont cvx]cvx bind def
}bind def
/MF{
findfont
[5 2 roll
0 3 1 roll
neg 0 0]makefont
dup setfont
[exch/setfont cvx]cvx bind def
}bind def
/level0 0 def
/RES 0 def
/PL 0 def
/LS 0 def
/MANUAL{
statusdict begin/manualfeed true store end
}bind def
/PLG{
gsave newpath clippath pathbbox grestore
exch pop add exch pop
}bind def
/BP{10 -52 translate
/level0 save def
1 setlinecap
1 setlinejoin
72 RES div dup scale
LS{
90 rotate
}{
0 PL translate
}ifelse
1 -1 scale
}bind def
/EP{
level0 restore
showpage
}bind def
/DA{
newpath arcn stroke
}bind def
/SN{
transform
.25 sub exch .25 sub exch
round .25 add exch round .25 add exch
itransform
}bind def
/DL{
SN
moveto
SN
lineto stroke
}bind def
/DC{
newpath 0 360 arc closepath
}bind def
/TM matrix def
/DE{
TM currentmatrix pop
translate scale newpath 0 0 .5 0 360 arc closepath
TM setmatrix
}bind def
/RC/rcurveto load def
/RL/rlineto load def
/ST/stroke load def
/MT/moveto load def
/CL/closepath load def
/Fr{
setrgbcolor fill
}bind def
/Fk{
setcmykcolor fill
}bind def
/Fg{
setgray fill
}bind def
/FL/fill load def
/LW/setlinewidth load def
/Cr/setrgbcolor load def
/Ck/setcmykcolor load def
/Cg/setgray load def
/RE{
findfont
dup maxlength 1 index/FontName known not{1 add}if dict begin
{
1 index/FID ne{def}{pop pop}ifelse
}forall
/Encoding exch def
dup/FontName exch def
currentdict end definefont pop
}bind def
/DEFS 0 def
/EBEGIN{
moveto
DEFS begin
}bind def
/EEND/end load def
/CNT 0 def
/level1 0 def
/PBEGIN{
/level1 save def
translate
div 3 1 roll div exch scale
neg exch neg exch translate
0 setgray
0 setlinecap
1 setlinewidth
0 setlinejoin
10 setmiterlimit
[]0 setdash
/setstrokeadjust where{
pop
false setstrokeadjust
}if
/setoverprint where{
pop
false setoverprint
}if
newpath
/CNT countdictstack def
userdict begin
/showpage{}def
}bind def
/PEND{
clear
countdictstack CNT sub{end}repeat
level1 restore
}bind def
end def
/setpacking where{
pop
setpacking
}if
%%EndResource
%%IncludeResource: font Courier-Bold
%%IncludeResource: font Times-Bold
%%IncludeResource: font Times-Roman
%%IncludeResource: font Times-Italic
%%IncludeResource: font Courier
grops begin/DEFS 1 dict def DEFS begin/u{.001 mul}bind def end/RES 72
def/PL 841.89 def/LS false def/ENC0[/asciicircum/asciitilde/Scaron
/Zcaron/scaron/zcaron/Ydieresis/trademark/quotesingle/Euro/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/space/exclam/quotedbl/numbersign/dollar/percent
/ampersand/quoteright/parenleft/parenright/asterisk/plus/comma/hyphen
/period/slash/zero/one/two/three/four/five/six/seven/eight/nine/colon
/semicolon/less/equal/greater/question/at/A/B/C/D/E/F/G/H/I/J/K/L/M/N/O
/P/Q/R/S/T/U/V/W/X/Y/Z/bracketleft/backslash/bracketright/circumflex
/underscore/quoteleft/a/b/c/d/e/f/g/h/i/j/k/l/m/n/o/p/q/r/s/t/u/v/w/x/y
/z/braceleft/bar/braceright/tilde/.notdef/quotesinglbase/guillemotleft
/guillemotright/bullet/florin/fraction/perthousand/dagger/daggerdbl
/endash/emdash/ff/fi/fl/ffi/ffl/dotlessi/dotlessj/grave/hungarumlaut
/dotaccent/breve/caron/ring/ogonek/quotedblleft/quotedblright/oe/lslash
/quotedblbase/OE/Lslash/.notdef/exclamdown/cent/sterling/currency/yen
/brokenbar/section/dieresis/copyright/ordfeminine/guilsinglleft
/logicalnot/minus/registered/macron/degree/plusminus/twosuperior
/threesuperior/acute/mu/paragraph/periodcentered/cedilla/onesuperior
/ordmasculine/guilsinglright/onequarter/onehalf/threequarters
/questiondown/Agrave/Aacute/Acircumflex/Atilde/Adieresis/Aring/AE
/Ccedilla/Egrave/Eacute/Ecircumflex/Edieresis/Igrave/Iacute/Icircumflex
/Idieresis/Eth/Ntilde/Ograve/Oacute/Ocircumflex/Otilde/Odieresis
/multiply/Oslash/Ugrave/Uacute/Ucircumflex/Udieresis/Yacute/Thorn
/germandbls/agrave/aacute/acircumflex/atilde/adieresis/aring/ae/ccedilla
/egrave/eacute/ecircumflex/edieresis/igrave/iacute/icircumflex/idieresis
/eth/ntilde/ograve/oacute/ocircumflex/otilde/odieresis/divide/oslash
/ugrave/uacute/ucircumflex/udieresis/yacute/thorn/ydieresis]def
/Courier@0 ENC0/Courier RE/Times-Italic@0 ENC0/Times-Italic RE
/Times-Roman@0 ENC0/Times-Roman RE/Times-Bold@0 ENC0/Times-Bold RE
/Courier-Bold@0 ENC0/Courier-Bold RE
%%EndProlog
%%Page: 1 1
%%BeginPageSetup
BP
%%EndPageSetup
/F0 10/Courier-Bold@0 SF(Internet Engineering Task Force)72 85 Q
(Sally Floyd)179.998 E 317.998(INTERNET-DRAFT ICIR)72 98 R 167.998
(draft-ietf-dccp-tfrc-voip-07.ps Eddie)72 111 R(Kohler)6 E
(Expires: May 2007)72 124 Q(UCLA)305.998 E(20 November 2006)407.998 137
Q/F1 14/Times-Bold@0 SF(TCP Friendly Rate Contr)178.086 162 Q
(ol \(TFRC\):)-.252 E(the Small-P)198.379 175 Q(ack)-.14 E(et \(SP\) V)
-.14 E(ariant)-1.288 E/F2 11/Times-Bold@0 SF(Status of this Memo)72 207
Q/F3 11/Times-Roman@0 SF
(By submitting this Internet-Draft, each author represents that an)100.8
223.6 Q 2.75(ya)-.165 G(pplicable patent or other)-2.75 E
(IPR claims of which he or she is a)100.8 236.6 Q -.11(wa)-.165 G(re ha)
.11 E .33 -.165(ve b)-.22 H(een or will be disclosed, and an).165 E 2.75
(yo)-.165 G 2.75(fw)-2.75 G(hich he)-2.75 E(or she becomes a)100.8 249.6
Q -.11(wa)-.165 G
(re will be disclosed, in accordance with Section 6 of BCP 79.).11 E
(Internet-Drafts are w)100.8 266.2 Q
(orking documents of the Internet Engineering T)-.11 E(ask F)-.88 E
(orce \(IETF\), its)-.165 E(areas, and its w)100.8 279.2 Q
(orking groups.)-.11 E(Note that other groups may also distrib)5.5 E
(ute w)-.22 E(orking)-.11 E(documents as Internet-Drafts.)100.8 292.2 Q
(Internet-Drafts are draft documents v)100.8 308.8 Q
(alid for a maximum of six months and may be)-.275 E
(updated, replaced, or obsoleted by other documents at an)100.8 321.8 Q
2.75(yt)-.165 G 2.75(ime. It)-2.75 F(is inappropriate to use)2.75 E
(Internet-Drafts as reference material or to cite them other than as "w)
100.8 334.8 Q(ork in progress.")-.11 E
(The list of current Internet-Drafts can be accessed at http://www)100.8
351.4 Q(.ietf.or)-.715 E(g/ietf/1id-)-.198 E(abstracts.txt.)100.8 364.4
Q(The list of Internet-Draft Shado)100.8 381 Q 2.75(wD)-.275 G
(irectories can be accessed at)-2.75 E(http://www)100.8 394 Q(.ietf.or)
-.715 E(g/shado)-.198 E -.715(w.)-.275 G(html.).715 E
(This Internet-Draft will e)100.8 410.6 Q(xpire on May 2007.)-.165 E F2
(Abstract)72 436.6 Q F3
(This document proposes a mechanism for further e)100.8 453.2 Q
(xperimentation, b)-.165 E(ut not for widespread)-.22 E(deplo)100.8
466.2 Q(yment at this time in the global Internet.)-.11 E(TCP-Friendly \
Rate Control \(TFRC\) is a congestion control mechanism for unicast \
\215o)100.8 482.8 Q(ws)-.275 E(operating in a best-ef)100.8 495.8 Q
(fort Internet en)-.275 E(vironment [RFC3448]. TFRC w)-.44 E
(as intended for)-.11 E(applications that use a \214x)100.8 508.8 Q
(ed pack)-.165 E(et size, and w)-.11 E(as designed to be reasonably f)
-.11 E(air when)-.11 E
(competing for bandwidth with TCP connections using the same pack)100.8
521.8 Q(et size.)-.11 E(This)5.5 E(document proposes TFRC-SP)100.8 534.8
Q 2.75(,aS)-1.221 G(mall-P)-2.75 E(ack)-.165 E(et \(SP\) v)-.11 E
(ariant of TFRC, that is designed for)-.275 E
(applications that send small pack)100.8 547.8 Q 2.75(ets. The)-.11 F
(design goal for TFRC-SP is to achie)2.75 E .33 -.165(ve t)-.275 H
(he same).165 E(bandwidth in bps \(bits per second\) as a TCP \215o)
100.8 560.8 Q 2.75(wu)-.275 G(sing pack)-2.75 E
(ets of up to 1500 bytes.)-.11 E(TFRC-SP enforces a minimum interv)100.8
573.8 Q(al of 10 ms between data pack)-.275 E(ets, to pre)-.11 E -.165
(ve)-.275 G(nt a single).165 E<8d6f>100.8 586.8 Q 2.75(wf)-.275 G
(rom sending small pack)-2.75 E(ets arbitrarily frequently)-.11 E(.)
-.715 E(Flo)100.8 603.4 Q(ws using TFRC-SP compete reasonably f)-.275 E
(airly with lar)-.11 E(ge-pack)-.198 E(et TCP and TFRC \215o)-.11 E(ws)
-.275 E(in en)100.8 616.4 Q(vironments where lar)-.44 E(ge-pack)-.198 E
(et \215o)-.11 E(ws and small-pack)-.275 E(et \215o)-.11 E(ws e)-.275 E
(xperience similar pack)-.165 E(et)-.11 E(drop rates.)100.8 629.4 Q(Ho)
5.5 E(we)-.275 E -.165(ve)-.275 G .88 -.44(r, i).165 H 2.75(ne).44 G
-.44(nv)-2.75 G(ironments where small-pack).44 E(et \215o)-.11 E(ws e)
-.275 E(xperience lo)-.165 E(wer pack)-.275 E(et)-.11 E
(drop rates than lar)100.8 642.4 Q(ge-pack)-.198 E(et \215o)-.11 E
(ws \(e.g., with Drop-T)-.275 E(ail queues in units of bytes\), TFRC-SP)
-.88 E(can recei)100.8 655.4 Q .33 -.165(ve c)-.275 H
(onsiderably more than its share of the bandwidth.).165 E(Flo)72 733 Q
(yd/K)-.11 E 333.671(ohler [P)-.385 F(age 1])-.165 E 0 Cg EP
%%Page: 2 2
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E .396
-.198(TO B)100.8 85 T 2.75(ED).198 G(ELETED BY THE RFC EDIT)-2.75 E
(OR UPON PUBLICA)-.198 E(TION:)-1.221 E
(Changes from draft-ietf-dccp-tfrc-v)103.55 98 Q(oip-06.txt:)-.22 E 2.75
(*F)106.3 111 S(ix)-2.75 E
(ed nits introduced in the last round of editing.)-.165 E
(Feedback from Gorry F)111.8 124 Q(airhurst.)-.165 E 2.75(*F)106.3 137 S
(ix)-2.75 E(ed T)-.165 E(ables 1-4, in terms of taking account of pack)
-.88 E(et)-.11 E(header sizes for 536-byte pack)111.8 150 Q 2.75
(ets. Feedback)-.11 F(from Ladan)2.75 E(Gharai.)111.8 163 Q 2.75(*A)
106.3 176 S(dded a paragraph about apps g)-2.75 E(aming about the pack)
-.055 E(et size,)-.11 E
(in Section 6 on "TFRC-SP with Applications that Modify)111.8 189 Q
(the P)111.8 202 Q(ack)-.165 E(et Size".)-.11 E(Feedback from Gorry F)
5.5 E(airhurst.)-.165 E 2.75(*A)106.3 215 S
(dded a paragraph that if the endpoint kno)-2.75 E(ws the MSS of)-.275 E
(the path, the endpoint SHOULD use that for the nominal)111.8 228 Q(se)
111.8 241 Q(gment size.)-.165 E(From discussions with Gorry F)5.5 E
(airhurst and)-.165 E(Richard Nelson.)111.8 254 Q 2.75(*A)106.3 267 S
(dded se)-2.75 E -.165(ve)-.275 G(ral paragrahs about TFRC-SP o).165 E
-.165(ve)-.165 G 2.75(rp).165 G(aths)-2.75 E(with small MTUs.)111.8 280
Q(Feedback from Gorry F)5.5 E(airhurst and)-.165 E(Richard Nelson.)111.8
293 Q 2.75(*A)106.3 306 S
(dded pseudocode to Section 3 on "TFRC-SP Congestion Control")-2.75 E
(for the change of not using the current interv)111.8 319 Q(al in)-.275
E(estimating the loss e)111.8 332 Q -.165(ve)-.275 G
(nt rate if the current interv).165 E(al)-.275 E(is short.)111.8 345 Q
(Feedback from Ladan Gharai.)5.5 E(Changes from draft-ietf-dccp-tfrc-v)
103.55 371 Q(oip-05.txt:)-.22 E 2.75(*F)106.3 384 S
(eedback from Gorry F)-2.75 E(airhurst:)-.165 E 2.75(-S)111.8 397 S
(mall editing changes, rephrasing, and b)-2.75 E(ug \214x)-.22 E(es.)
-.165 E 2.75(-E)111.8 410 S
(xplicitly stated that assuming a 40-byte header)-2.75 E(is OK e)117.3
423 Q -.165(ve)-.275 G 2.75(ni).165 G 2.75(fI)-2.75 G(Pv6 is used.)-2.75
E 2.75(-M)111.8 436 S -.165(ove)-2.75 G 2.75(dm).165 G
(ost of Section 7 on Simulations to)-2.75 E(the appendix.)117.3 449 Q
2.75(-A)111.8 462 S(dded a subsection to Section 8 on "F)-2.75 E
(airness with)-.165 E(dif)117.3 475 Q(ferent pack)-.275 E
(et header sizes")-.11 E 2.75(-A)111.8 488 S
(dd a paragraph about TFRC-SP and TFRC-PS.)-2.75 E 2.75(-As)111.8 501 S
(udden step-change in the pack)-2.75 E(et size? -)-.11 E
(no change made to algorithm.)122.8 514 Q 2.75(-C)111.8 527 S
(lari\214ed te)-2.75 E(xt on impact of a reduced TCP MSS.)-.165 E 2.75
(*F)106.3 540 S(eedback from Lars Eggert:)-2.75 E 2.75(-S)111.8 553 S
(mall editing changes, rephrasing, and b)-2.75 E(ug \214x)-.22 E(es.)
-.165 E 2.75(*F)106.3 566 S(eedback from Mark Handle)-2.75 E(y:)-.165 E
2.75(-N)111.8 579 S 2.75(oc)-2.75 G(hange made to algorithm, b)-2.75 E
(ut added Appendix C)-.22 E(to discuss the issue.)117.3 592 Q 2.75(*F)
106.3 605 S(eedback from Ladan Gharai:)-2.75 E 2.75(-A)111.8 618 S
(dded the restriction that the most recent loss interv)-2.75 E(al)-.275
E(is not included in the calculation of the a)117.3 631 Q -.165(ve)-.22
G(rage).165 E(loss interv)117.3 644 Q(al if the most recent loss interv)
-.275 E(al is short.)-.275 E 2.75(-C)111.8 657 S(orrected use of "pack)
-2.75 E(et" and "se)-.11 E(gment" in a fe)-.165 E 2.75(wp)-.275 G
(laces.)-2.75 E(Changes from draft-ietf-dccp-tfrc-v)103.55 683 Q
(oip-04.txt:)-.22 E 2.75(*A)106.3 696 S(dded tables sho)-2.75 E
(wing the response function for TCP)-.275 E 2.75(,T)-1.221 G(FRC,)-2.75
E(Flo)72 733 Q(yd/K)-.11 E 333.671(ohler [P)-.385 F(age 2])-.165 E 0 Cg
EP
%%Page: 3 3
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E
(and TFRC-SP)111.8 85 Q 2.75(,f)-1.221 G(or a range of pack)-2.75 E
(et sizes, for both)-.11 E(pack)111.8 98 Q(et and byte drop rates.)-.11
E(In response to feedback)5.5 E(from Magnus W)111.8 111 Q(esterlund.)
-.88 E 2.75(*A)106.3 124 S(long with response function, added that TCP')
-2.75 E 2.75(ss)-.605 G(ending)-2.75 E(rate v)111.8 137 Q
(aries linearly with pack)-.275 E(et size.)-.11 E(From a)5.5 E
(suggestion by Magnus W)111.8 150 Q(esterlund.)-.88 E 2.75(*A)106.3 163
S(dded a sentence saying that TCP has a range of sender)-2.75 E
(algorithms for setting the R)111.8 176 Q -.198(TO)-.66 G(.).198 E 2.75
(*D)106.3 189 S(eleted the sentence equating TFRC-SP with TFRC-PS)-2.75
E(referred to in RFC 3448.)111.8 202 Q(From a suggestion)5.5 E
(by Colin Perkins.)111.8 215 Q 2.75(*A)106.3 228 S
(dded that wireless links sometimes are less lik)-2.75 E(ely to)-.11 E
(drop small pack)111.8 241 Q 2.75(ets. Reported)-.11 F
(from Pete Sholander)2.75 E(.)-.605 E 2.75(*A)106.3 254 S
(dded simulations to the end of Section 7.3 comparing)-2.75 E(the ef)
111.8 267 Q(fects of TFRC and of TFRC-SP)-.275 E 2.75(,f)-1.221 G
(or an en)-2.75 E(vironment)-.44 E(with a Drop-T)111.8 280 Q
(ail queue in bytes, sho)-.88 E(wing the possible)-.275 E(ne)111.8 293 Q
-.055(ga)-.165 G(ti).055 E .33 -.165(ve c)-.275 H
(onsequences of TFRC-SP).165 E 5.5(.I)-1.221 G 2.75(nr)-5.5 G
(esponse to email)-2.75 E(from Magnus W)111.8 306 Q(esterlund.)-.88 E
2.75(*A)106.3 319 S(dded an e)-2.75 E(xplanation for the Adapti)-.165 E
.33 -.165(ve R)-.275 H(ED simulation).165 E(with pack)111.8 332 Q
(et drop rates greater than 50%. In response)-.11 E
(to email from Magnus W)111.8 345 Q(esterlund.)-.88 E 2.75(*A)106.3 358
S(dded a Conclusions section, with a sentence that a)-2.75 E
(separate document will be used to specify an)111.8 371 Q -.165(ex)111.8
384 S(perimental CCID based on TFRC-SP).165 E 5.5(.I)-1.221 G 2.75(nr)
-5.5 G(esponse)-2.75 E(to feedback during W)111.8 397 Q
(orking Group Last Call.)-.88 E 2.75(*A)106.3 410 S
(dded a paragraph about "Initializing the Loss History)-2.75 E
(after the First Loss Ev)111.8 423 Q(ent" in TFRC-SP)-.165 E(.)-1.221 E
(Changes from draft-ietf-dccp-tfrc-v)103.55 449 Q(oip-03.txt:)-.22 E
2.75(*A)106.3 462 S(dded a paragraph saying that this is intended for)
-2.75 E(Experimental, for further e)111.8 475 Q(xperimentation and not)
-.165 E(for widespread deplo)111.8 488 Q(yment.)-.11 E 2.75(*E)106.3 501
S(diting of abstract so that it still \214ts the 25-line)-2.75 E(limit.)
111.8 514 Q(Changes from draft-ietf-dccp-tfrc-v)103.55 540 Q
(oip-02.txt:)-.22 E 2.75(*C)106.3 553 S(hanged name from "V)-2.75 E
(oIP v)-1.419 E(ariant of TFRC" to "TFRC-SP".)-.275 E 2.75(*A)106.3 566
S(dded Section 4.5 on "The Nominal P)-2.75 E(ack)-.165 E
(et Size", discussing)-.11 E(possible dif)114.55 579 Q(ferences in pack)
-.275 E(et drop rates between small)-.11 E(and lar)114.55 592 Q(ge pack)
-.198 E(ets.)-.11 E 2.75(*A)106.3 605 S(dded te)-2.75 E
(xt to Section 5 on "A Comparison with RFC 3714".)-.165 E 2.75(*A)106.3
618 S(dded te)-2.75 E
(xt to Section 6 on "TFRC-SP with Applications that)-.165 E
(Modify the P)117.3 631 Q(ack)-.165 E(et Size")-.11 E 2.75(*A)106.3 644
S(dded simulations with small-pack)-2.75 E(et TCP \215o)-.11 E(ws.)-.275
E 2.75(*A)106.3 657 S(dded a Security Considerations section.)-2.75 E
2.75(*M)106.3 670 S(inor editing.)-2.75 E
(Changes from draft-ietf-dccp-tfrc-v)103.55 696 Q(oip-01.txt:)-.22 E
(Flo)72 733 Q(yd/K)-.11 E 333.671(ohler [P)-.385 F(age 3])-.165 E 0 Cg
EP
%%Page: 4 4
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E 2.75(*A)
106.3 85 S(dded modi\214ed algorithm for calculating the loss e)-2.75 E
-.165(ve)-.275 G(nt rate,).165 E(for short interv)117.3 98 Q
(als with multiple pack)-.275 E(et drops.)-.11 E 2.75(*M)106.3 111 S
-.165(ove)-2.75 G 2.75(dF).165 G(aster Restart to a separate document.)
-2.915 E 2.75(*A)106.3 124 S
(dded simulations with a con\214gured byte drop rate.)-2.75 E 2.75(*A)
106.3 137 S(dded man)-2.75 E 2.75(ym)-.165 G
(ore simulations, including Drop-T)-2.75 E(ail with a queue)-.88 E
(in bytes.)111.8 150 Q 2.75(*A)106.3 163 S(dded a discussion of unf)
-2.75 E(airness for Drop-T)-.11 E(ail with a queue)-.88 E(in bytes.)
111.8 176 Q(Changes from draft-ietf-dccp-tfrc-v)103.55 202 Q
(oip-00.txt:)-.22 E 2.75(*A)106.3 215 S(dded more simulations.)-2.75 E
2.75(*A)106.3 228 S(dded a Related W)-2.75 E(ork section.)-.88 E(Flo)72
733 Q(yd/K)-.11 E 333.671(ohler [P)-.385 F(age 4])-.165 E 0 Cg EP
%%Page: 5 5
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E/F1 11
/Times-Bold@0 SF -1.012(Ta)72 85 S(ble of Contents)1.012 E/F2 10
/Times-Roman@0 SF(1. Con)100.8 110 Q -.15(ve)-.4 G(ntions).15 E F0 -2.25
(......................................................................\
............................................... ....)4.94 F F2(6)A
(2. Introduction)100.8 122 Q F0 -2.25(.................................\
.......................................................................\
.............. ....)2.76 F F2(6)A(3. TFRC-SP Congestion Control)100.8
134 Q F0 -2.25(........................................................\
.................................... ....)3.12 F F2(8)A
(4. TFRC-SP Discussion)100.8 146 Q F0 -1.75(...........................\
.......................................................................\
...... ..)4.85 F F2(10)A
(4.1. Response Functions and Throughput Equations)115.8 158 Q F0 -1.75
(........................................................... ..)2.51 F
F2(10)A(4.2. Accounting for Header Size)115.8 170 Q F0 -1.75(..........\
.......................................................................\
...... ..)3.04 F F2(14)A(4.3. The TFRC-SP Min Interv)115.8 182 Q(al)-.25
E F0 -1.75(............................................................\
........................... ..)4.92 F F2(14)A(4.4. Counting P)115.8 194
Q(ack)-.15 E(et Losses)-.1 E F0 -1.75(.................................\
............................................................ ..)2.59 F
F2(15)A(4.5. The Nominal P)115.8 206 Q(ack)-.15 E(et Size)-.1 E F0 -1.75
(......................................................................\
.................... ..)5.02 F F2(16)A(4.5.1. P)140.8 218 Q(ack)-.15 E
(et Size and P)-.1 E(ack)-.15 E(et Drop Rates)-.1 E F0 -1.75
(................................................................. ..)
3.2 F F2(16)A(4.5.2. Fragmentation and the P)140.8 230 Q(ath MTU)-.15 E
F0 -1.75
(................................................................... ..)
2.61 F F2(17)A(4.5.3. The Nominal Se)140.8 242 Q(gment Size and the P)
-.15 E(ath MTU)-.15 E F0 -1.75
(................................................ ..)3.07 F F2(17)A
(4.6. The Loss Interv)115.8 254 Q(al Length for Short Loss Interv)-.25 E
(als)-.25 E F0 -1.75
(....................................................... ..)4.87 F F2
(18)A(5. A Comparison with RFC 3714)100.8 266 Q F0 -1.75(..............\
.......................................................................\
...... ..)3.66 F F2(18)A(6. TFRC-SP with Applications that Modify the P)
100.8 278 Q(ack)-.15 E(et Size)-.1 E F0 -1.75
(..................................................... ..)4.25 F F2(19)A
(7. Simulations)100.8 290 Q F0 -1.75(..................................\
.......................................................................\
............. ..)4.41 F F2(20)A(8. General Discussion)100.8 302 Q F0
-1.75(.................................................................\
.......................................... ..)4.41 F F2(20)A
(9. Security Considerations)100.8 314 Q F0 -1.75(......................\
.......................................................................\
........ ..)3.12 F F2(21)A(10. IAN)100.8 326 Q 2.5(AC)-.35 G
(onsiderations)-2.5 E F0 -1.75(........................................\
.............................................................. ..)4.06 F
F2(21)A(11. Conclusions)100.8 338 Q F0 -1.75(..........................\
.......................................................................\
................... ..)3.25 F F2(21)A(12. Thanks)100.8 350 Q F0 -1.75(.\
.......................................................................\
................................................... ..)4.01 F F2(22)A
(A. Appendix: Related W)100.8 362 Q(ork on Small-P)-.8 E(ack)-.15 E
(et V)-.1 E(ariants of TFRC)-1.11 E F0 -1.75
(............................................... ..)3.8 F F2(22)A
(B. Simulation Results)100.8 374 Q F0 -1.75(...........................\
.......................................................................\
......... ..)4.93 F F2(23)A(B.1. Simulations with Con\214gured P)115.8
386 Q(ack)-.15 E(et Drop Rates)-.1 E F0 -1.75
(......................................................... ..)3.53 F F2
(23)A(B.2. Simulations with Con\214gured Byte Drop Rates)115.8 398 Q F0
-1.75(............................................................ ..)
2.8 F F2(26)A(B.3. P)115.8 410 Q(ack)-.15 E(et Dropping Beha)-.1 E
(vior at Routers with Drop-T)-.2 E(ail Queues)-.8 E F0 -1.75
(...................................... ..)4.04 F F2(28)A(B.4. P)115.8
422 Q(ack)-.15 E(et Dropping Beha)-.1 E(vior at Routers with A)-.2 E(QM)
-.55 E F0 -1.75
(........................................................ ..)3.44 F F2
(32)A(C. Appendix: Exploring Possible Oscillations in the Loss Ev)100.8
434 Q(ent Rate)-.15 E F0 -1.75
(....................................... ..)4.04 F F2(36)A
(D. Appendix: A Discussion of P)100.8 446 Q(ack)-.15 E(et Size and P)-.1
E(ack)-.15 E(et Dropping)-.1 E F0 -1.75
(.............................................. ..)3.51 F F2(37)A
(Normati)100.8 458 Q .3 -.15(ve R)-.25 H(eferences).15 E F0 -1.75(.....\
.......................................................................\
............................... ..)3.16 F F2(37)A(Informati)100.8 470 Q
.3 -.15(ve R)-.25 H(eferences).15 E F0 -1.75(..........................\
.......................................................................\
........ ..)4.22 F F2(37)A(Authors' Addresses)100.8 482 Q F0 -1.75(....\
.......................................................................\
................................... ..)5.05 F F2(38)A(Full Cop)100.8 494
Q(yright Statement)-.1 E F0 -1.75(.....................................\
................................................................. ..)
4.62 F F2(38)A(Intellectual Property)100.8 506 Q F0 -1.75(.............\
.......................................................................\
......................... ..)5.02 F F2(39)A F0(Flo)72 733 Q(yd/K)-.11 E
333.671(ohler [P)-.385 F(age 5])-.165 E 0 Cg EP
%%Page: 6 6
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E/F1 11
/Times-Bold@0 SF(1.)72 85 Q/F2 14/Times-Bold@0 SF(Con)5.5 E -.14(ve)-.56
G(ntions).14 E F0(The k)100.8 101.6 Q .33 -.165(ey w)-.11 H
(ords "MUST", "MUST NO).055 E(T", "REQ)-.44 E
(UIRED", "SHALL", "SHALL NO)-.11 E(T",)-.44 E("SHOULD", "SHOULD NO)100.8
114.6 Q(T", "RECOMMENDED", "MA)-.44 E(Y", and "OPTION)-1.155 E
(AL" in this)-.385 E
(document are to be interpreted as described in [RFC2119].)100.8 127.6 Q
F1(2.)72 153.6 Q F2(Intr)5.5 E(oduction)-.252 E F0
(This document speci\214es TFRC-SP)100.8 170.2 Q 2.75(,a)-1.221 G 2.75
(ne)-2.75 G(xperimental, Small-P)-2.915 E(ack)-.165 E(et v)-.11 E
(ariant of TCP-friendly)-.275 E(rate control \(TFRC\) [RFC3448].)100.8
183.2 Q(TFRC w)100.8 199.8 Q(as designed to be reasonably f)-.11 E
(air when competing for bandwidth with TCP \215o)-.11 E(ws,)-.275 E -.22
(bu)100.8 212.8 S 2.75(tt).22 G 2.75(oa)-2.75 G -.22(vo)-2.97 G
(id the abrupt changes in the sending rate characteristic of TCP').22 E
2.75(sc)-.605 G(ongestion)-2.75 E(control mechanisms.)100.8 225.8 Q
(TFRC is intended for applications such as streaming media)5.5 E
(applications where a relati)100.8 238.8 Q -.165(ve)-.275 G
(ly smooth sending rate is of importance.).165 E(Con)5.5 E -.165(ve)-.44
G(ntional TFRC).165 E(measures loss rates by estimating the loss e)100.8
251.8 Q -.165(ve)-.275 G(nt ratio as described in [RFC3448], and uses)
.165 E(this loss e)100.8 264.8 Q -.165(ve)-.275 G
(nt rate to determine the sending rate in pack).165 E
(ets per round-trip time.)-.11 E(This has)5.5 E
(consequences for the rate that a TFRC \215o)100.8 277.8 Q 2.75(wc)-.275
G(an achie)-2.75 E .33 -.165(ve w)-.275 H(hen sharing a bottleneck with)
.165 E(lar)100.8 290.8 Q(ge-pack)-.198 E(et TCP \215o)-.11 E 2.75
(ws. In)-.275 F(particular)2.75 E 2.75(,al)-.44 G -.275(ow)-2.75 G
(-bandwidth, small-pack).275 E(et TFRC \215o)-.11 E 2.75(ws)-.275 G
(haring a)-2.75 E(bottleneck with high-bandwidth, lar)100.8 303.8 Q
(ge-pack)-.198 E(et TCP \215o)-.11 E(ws may be forced to slo)-.275 E
2.75(wd)-.275 G -.275(ow)-2.75 G(n, e).275 E -.165(ve)-.275 G(n).165 E
(though the TFRC application')100.8 316.8 Q 2.75(sn)-.605 G
(ominal rate in bytes per second is less than the rate)-2.75 E(achie)
100.8 329.8 Q -.165(ve)-.275 G 2.75(db).165 G 2.75(yt)-2.75 G
(he TCP \215o)-2.75 E 2.75(ws. Intuiti)-.275 F -.165(ve)-.275 G(ly).165
E 2.75(,t)-.715 G(his w)-2.75 E(ould be "f)-.11 E(air" only if the netw)
-.11 E(ork limitation)-.11 E -.11(wa)100.8 342.8 S 2.75(si).11 G 2.75
(np)-2.75 G(ack)-2.75 E(ets per second \(such as a routing lookup\), ra\
ther than bytes per second \(such as)-.11 E(link bandwidth\).)100.8
355.8 Q(Con)5.5 E -.165(ve)-.44 G(ntional wisdom is that man).165 E 2.75
(yo)-.165 G 2.75(ft)-2.75 G(he netw)-2.75 E(ork limitations in today')
-.11 E(s)-.605 E(Internet are in bytes per second, e)100.8 368.8 Q -.165
(ve)-.275 G 2.75(nt).165 G(hough the netw)-2.75 E
(ork limitations of the future might)-.11 E(mo)100.8 381.8 Q .33 -.165
(ve b)-.165 H(ack to).165 E -.11(wa)-.275 G(rds limitations in pack).11
E(ets per second.)-.11 E(TFRC-SP is intended for \215o)100.8 398.4 Q
(ws that need to send frequent small pack)-.275 E(ets, with less than)
-.11 E(1500 bytes per pack)100.8 411.4 Q
(et, limited by a minimum interv)-.11 E(al between pack)-.275 E
(ets of 10 ms.)-.11 E(It will)5.5 E
(better support applications that do not w)100.8 424.4 Q
(ant their sending rates in bytes per second to suf)-.11 E(fer)-.275 E
(from their use of small pack)100.8 437.4 Q 2.75(ets. This)-.11 F -.275
(va)2.75 G(riant is restricted to applications that send pack).275 E
(ets)-.11 E(no more than once e)100.8 450.4 Q -.165(ve)-.275 G
(ry 10 ms \(the Min Interv).165 E(al or minimum interv)-.275 E 2.75
(al\). Gi)-.275 F -.165(ve)-.275 G 2.75(nt).165 G(his)-2.75 E
(restriction, TFRC-SP ef)100.8 463.4 Q(fecti)-.275 E -.165(ve)-.275 G
(ly calculates the TFRC f).165 E(air rate as if the bottleneck)-.11 E
(restriction w)100.8 476.4 Q(as in bytes per second.)-.11 E
(Applications using TFRC-SP could ha)5.5 E .33 -.165(ve a \214)-.22 H
-.165(xe).165 G 2.75(do).165 G(r)-2.75 E(naturally-v)100.8 489.4 Q
(arying pack)-.275 E(et size, or could v)-.11 E(ary their pack)-.275 E
(et size in response to congestion.)-.11 E
(Applications that are not willing to be limited by a minimum interv)
100.8 502.4 Q(al of 10 ms. between)-.275 E(pack)100.8 515.4 Q
(ets, or that w)-.11 E(ant to send pack)-.11 E(ets lar)-.11 E
(ger than 1500 bytes, should not use TFRC-SP)-.198 E(.)-1.221 E(Ho)100.8
528.4 Q(we)-.275 E -.165(ve)-.275 G .88 -.44(r, f).165 H
(or applications with a minimum interv).44 E
(al of at least 10 ms. between pack)-.275 E(ets and)-.11 E
(with data pack)100.8 541.4 Q(ets of at most 1500 bytes, the performanc\
e of TFRC-SP should be at least as)-.11 E(good as that from TFRC.)100.8
554.4 Q(RFC 3448, the protocol speci\214cation for TFRC, stated that TF\
RC-PS \(for TFRC-)100.8 571 Q -.165(Pa)100.8 584 S(ck).165 E
(etSize\), a v)-.11 E(ariant of TFRC for applications that ha)-.275 E
.33 -.165(ve a \214)-.22 H -.165(xe).165 G 2.75(ds).165 G(ending rate b)
-2.75 E(ut v)-.22 E(ary their)-.275 E(pack)100.8 597 Q
(et size in response to congestion, w)-.11 E
(ould be speci\214ed in a later document.)-.11 E(This)5.5 E
(document instead speci\214es TFRC-SP)100.8 610 Q 2.75(,av)-1.221 G
(ariant of TFRC designed for applications that)-3.025 E(send small pack)
100.8 623 Q(ets, where applications could either ha)-.11 E .33 -.165
(ve a \214)-.22 H -.165(xe).165 G 2.75(do).165 G 2.75(rv)-2.75 G
(arying pack)-3.025 E(et size or)-.11 E(could adapt their pack)100.8 636
Q(et size in response to congestion.)-.11 E(Ho)5.5 E(we)-.275 E -.165
(ve)-.275 G .88 -.44(r, a).165 H 2.75(sd).44 G(iscussed in Section 6)
-2.75 E(of this document, there are man)100.8 649 Q 2.75(yq)-.165 G
(uestions about ho)-2.75 E 2.75(ws)-.275 G(uch an adapti)-2.75 E .33
-.165(ve a)-.275 H(pplication w).165 E(ould)-.11 E
(use TFRC-SP that are be)100.8 662 Q
(yond the scope of this document, and that w)-.165 E(ould need to be)
-.11 E(addressed in documents that are more application-speci\214c.)
100.8 675 Q(Flo)72 733 Q(yd/K)-.11 E 284.171(ohler Section)-.385 F 2.75
(2. [P)2.75 F(age 6])-.165 E 0 Cg EP
%%Page: 7 7
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E
(TFRC-SP is moti)100.8 85 Q -.275(va)-.275 G
(ted in part by the approach in RFC 3714, which ar).275 E
(gues that it is)-.198 E(acceptable for V)100.8 98 Q(oIP \215o)-1.419 E
(ws to assume that the netw)-.275 E
(ork limitation is in bytes per second)-.11 E(\(Bps\) rather in pack)
100.8 111 Q(ets per second \(pps\), and to ha)-.11 E .33 -.165(ve t)-.22
H(he same sending rate in bytes per).165 E(second as a TCP \215o)100.8
124 Q 2.75(ww)-.275 G(ith 1500-byte pack)-2.75 E(ets and the same pack)
-.11 E(et drop rate.)-.11 E(RFC 3714)5.5 E(states the follo)100.8 137 Q
(wing:)-.275 E("While the ideal w)129.6 153.6 Q(ould be to ha)-.11 E .33
-.165(ve a t)-.22 H(ransport protocol that is able to detect whether)
.165 E(the bottleneck links along the path are limited in Bps or in pps\
, and to respond)129.6 166.6 Q(appropriately when the limitation is in \
pps, such an ideal is hard to achie)129.6 179.6 Q -.165(ve)-.275 G 2.75
(.W).165 G(e)-3.63 E -.11(wo)129.6 192.6 S(uld not w).11 E
(ant to delay the deplo)-.11 E(yment of congestion control for telephon)
-.11 E 2.75(yt)-.165 G(raf)-2.75 E<8c63>-.275 E
(until such an ideal could be accomplished.)129.6 205.6 Q
(In addition, we note that the current TCP)5.5 E
(congestion control mechanisms are themselv)129.6 218.6 Q(es not v)-.165
E(ery ef)-.165 E(fecti)-.275 E .33 -.165(ve i)-.275 H 2.75(na).165 G
2.75(ne)-2.75 G -.44(nv)-2.75 G(ironment).44 E
(where there is a limitation along the re)129.6 231.6 Q -.165(ve)-.275 G
(rse path in pps.).165 E(While the TCP)5.5 E(mechanisms do pro)129.6
244.6 Q(vide an incenti)-.165 E .33 -.165(ve t)-.275 H 2.75(ou).165 G
(se lar)-2.75 E(ge data pack)-.198 E(ets, TCP does not include)-.11 E
(an)129.6 257.6 Q 2.75(ye)-.165 G -.275(ff)-2.75 G(ecti).275 E .33 -.165
(ve c)-.275 H(ongestion control mechanisms for the stream of small).165
E(ackno)129.6 270.6 Q(wledgement pack)-.275 E(ets on the re)-.11 E -.165
(ve)-.275 G(rse path.).165 E(Gi)5.5 E -.165(ve)-.275 G 2.75(nt).165 G
(he ar)-2.75 E(guments abo)-.198 E -.165(ve)-.165 G 2.75(,i).165 G 2.75
(ts)-2.75 G(eems)-2.75 E(acceptable to us to assume a netw)129.6 283.6 Q
(ork limitation in Bps rather than in pps in)-.11 E
(considering the minimum sending rate of telephon)129.6 296.6 Q 2.75(yt)
-.165 G(raf)-2.75 E(\214c.")-.275 E -.385(Tr)100.8 313.2 S(anslating th\
e discussion in [RFC3714] to the congestion control mechanisms of TFRC,\
 it).385 E(seems acceptable to standardize a v)100.8 326.2 Q
(ariant of TFRC that allo)-.275 E(ws lo)-.275 E(w-bandwidth \215o)-.275
E(ws)-.275 E(sending small pack)100.8 339.2 Q(ets to achie)-.11 E .33
-.165(ve a r)-.275 H(ough f).165 E(airness with TCP \215o)-.11 E
(ws in terms of the sending)-.275 E
(rate in Bps, rather than in terms of the sending rate in pps.)100.8
352.2 Q(This is accomplished by TFRC-)5.5 E(SP)100.8 365.2 Q 2.75(,as)
-1.221 G(mall modi\214cation to TFRC, as described belo)-2.75 E -.715
(w.)-.275 G(Maintaining incenti)100.8 381.8 Q -.165(ve)-.275 G 2.75(sf)
.165 G(or lar)-2.75 E(ge pack)-.198 E
(ets: Because the bottlenecks in the netw)-.11 E(ork in f)-.11 E
(act can)-.11 E(include limitations in pps as well as in Bps, we pay sp\
ecial attention to the potential)100.8 394.8 Q
(dangers of encouraging a lar)100.8 407.8 Q(ge deplo)-.198 E
(yment of best-ef)-.11 E(fort traf)-.275 E
(\214c in the Internet consisting)-.275 E(entirely of small pack)100.8
420.8 Q 2.75(ets. This)-.11 F
(is discussed in more detail in Section 4.3. In addition, as)2.75 E(ag)
100.8 433.8 Q(ain discussed in Section 4.3, TFRC-SP includes the limita\
tion of the Min Interv)-.055 E(al)-.275 E(between pack)100.8 446.8 Q
(ets of 10 ms.)-.11 E -.165(Pa)100.8 463.4 S(ck).165 E
(et drop rates as a function of pack)-.11 E
(et size: TFRC-SP essentially assumes that the small-)-.11 E(pack)100.8
476.4 Q(et TFRC-SP \215o)-.11 E 2.75(wr)-.275 G(ecei)-2.75 E -.165(ve)
-.275 G 2.75(sr).165 G(oughly the same pack)-2.75 E
(et drop rate as a lar)-.11 E(ge-pack)-.198 E(et TFRC)-.11 E
(or TCP \215o)100.8 489.4 Q 4.18 -.715(w. A)-.275 H 2.75(sw).715 G 2.75
(es)-2.75 G(ho)-2.75 E 1.43 -.715(w, t)-.275 H
(his assumption is not necessarily correct for all en).715 E(vironments)
-.44 E(in the Internet.)100.8 502.4 Q
(Initializing the Loss History after the First Loss Ev)100.8 519 Q
(ent: Section 6.3.1 of RFC 3448 speci\214es)-.165 E(that the TFRC recei)
100.8 532 Q -.165(ve)-.275 G 2.75(ri).165 G
(nitializes the loss history after the \214rst loss e)-2.75 E -.165(ve)
-.275 G(nt by calculating the).165 E(loss interv)100.8 545 Q(al that w)
-.275 E(ould be required to produce the recei)-.11 E .33 -.165(ve r)
-.275 H(ate measured o).165 E -.165(ve)-.165 G 2.75(rt).165 G(he most)
-2.75 E(recent round-trip time.)100.8 558 Q
(In calculating this loss interv)5.5 E(al, TFRC-SP uses the se)-.275 E
(gment size of)-.165 E(1460 bytes, rather than the actual se)100.8 571 Q
(gment size used in the connection.)-.165 E(Calculating the loss e)100.8
587.6 Q -.165(ve)-.275 G
(nt rate for TFRC-SP: TFRC-SP requires a modi\214cation in TFRC').165 E
(s)-.605 E(calculation of the loss e)100.8 600.6 Q -.165(ve)-.275 G
(nt rate, because a TFRC-SP connection can send man).165 E 2.75(ys)-.165
G(mall)-2.75 E(pack)100.8 613.6 Q
(ets when a standard TFRC or TCP connection w)-.11 E
(ould send a single lar)-.11 E(ge pack)-.198 E 2.75(et. It)-.11 F(is)
2.75 E(not possible for a standard TFRC or TCP connection to repeatedly\
 send multiple pack)100.8 626.6 Q(ets)-.11 E
(per round-trip time in the f)100.8 639.6 Q(ace of a high pack)-.11 E
(et drop rate.)-.11 E(As a result, TCP and standard)5.5 E
(TFRC only respond to a single loss e)100.8 652.6 Q -.165(ve)-.275 G
(nt per round-trip time, and are still able to detect).165 E
(and respond to increasingly hea)100.8 665.6 Q(vy pack)-.22 E
(et loss rates.)-.11 E(Ho)5.5 E(we)-.275 E -.165(ve)-.275 G .88 -.44
(r, i).165 H 2.75(nah).44 G(ighly-congested)-2.75 E(en)100.8 678.6 Q
(vironment, when a TCP connection might be sending, on a)-.44 E -.165
(ve)-.22 G(rage, one lar).165 E(ge pack)-.198 E(et per)-.11 E(round-tri\
p time, a corresponding TFRC-SP connection might be sending man)100.8
691.6 Q 2.75(ys)-.165 G(mall)-2.75 E(Flo)72 733 Q(yd/K)-.11 E 284.171
(ohler Section)-.385 F 2.75(2. [P)2.75 F(age 7])-.165 E 0 Cg EP
%%Page: 8 8
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E(pack)
100.8 85 Q(ets per round-trip time.)-.11 E
(As a result, in order to maintain f)5.5 E(airness with TCP)-.11 E 2.75
(,a)-1.221 G(nd to be)-2.75 E(able to detect changes in the de)100.8 98
Q(gree of congestion, TFRC-SP needs to be sensiti)-.165 E .33 -.165
(ve t)-.275 H 2.75(ot).165 G(he)-2.75 E(actual pack)100.8 111 Q
(et drop rate during periods of high congestion.)-.11 E
(This is discussed in more detail)5.5 E(in the section belo)100.8 124 Q
-.715(w.)-.275 G/F1 11/Times-Bold@0 SF(3.)72 150 Q/F2 14/Times-Bold@0 SF
(TFRC-SP Congestion Contr)5.5 E(ol)-.252 E F0
(TFRC uses the TCP throughput equation gi)100.8 166.6 Q -.165(ve)-.275 G
2.75(ni).165 G 2.75(nS)-2.75 G(ection 3.1 of [RFC3448], which gi)-2.75 E
-.165(ve)-.275 G(s).165 E(the allo)100.8 179.6 Q
(wed sending rate X in bytes per second as a function of the loss e)
-.275 E -.165(ve)-.275 G(nt rate, se).165 E(gment)-.165 E
(size, and round-trip time.)100.8 192.6 Q
([RFC3448] speci\214es that the se)5.5 E(gment size s used in the)-.165
E(throughput equation should be the se)100.8 205.6 Q
(gment size used by the application, or the estimated)-.165 E(mean se)
100.8 218.6 Q(gment size if there are v)-.165 E(ariations in the se)
-.275 E(gment size depending on the data.)-.165 E(This)5.5 E(gi)100.8
231.6 Q -.165(ve)-.275 G 2.75(sr).165 G(ough f)-2.75 E
(airness with TCP \215o)-.11 E(ws using the same se)-.275 E(gment size.)
-.165 E(TFRC-SP changes this beha)100.8 248.2 Q(vior in the follo)-.22 E
(wing w)-.275 E(ays.)-.11 E 12.65<8354>100.8 264.8 S(he nominal se)
-12.65 E(gment size: The nominal se)-.165 E
(gment size s is set to 1460 bytes.)-.165 E -.165(Fo)5.5 G(llo).165 E
(wing)-.275 E([RFC3714], this pro)117.3 277.8 Q(vides a goal of f)-.165
E(airness, in terms of the sending rate in bytes per)-.11 E
(second, with a TCP \215o)117.3 290.8 Q 2.75(ww)-.275 G
(ith 1460 bytes of application data per pack)-2.75 E(et b)-.11 E
(ut with the)-.22 E(same pack)117.3 303.8 Q(et drop rate.)-.11 E
(If the endpoint kno)5.5 E(ws the MTU \(Maximum T)-.275 E
(ransmission Unit\))-.385 E(of the path and the deri)117.3 316.8 Q -.165
(ve)-.275 G 2.75(dM).165 G(SS \(Maximum Se)-2.75 E
(gment Size\) is less than 1460 bytes,)-.165 E
(then the endpoint SHOULD set the nominal se)117.3 329.8 Q
(gment size s to MSS bytes.)-.165 E(In addition,)5.5 E
(if the endpoint kno)117.3 342.8 Q
(ws the MTU of the path and the resulting MSS is less than 536)-.275 E
(bytes, then the endpoint MUST set the nominal se)117.3 355.8 Q
(gment size s to MSS bytes.)-.165 E(Ho)117.3 372.4 Q(we)-.275 E -.165
(ve)-.275 G .88 -.44(r, t).165 H(his document does not require that TFR\
C-SP endpoints determine the path).44 E 2.75(MTU. While)117.3 385.4 R
(most paths allo)2.75 E 2.75(wa)-.275 G 2.75(nM)-2.75 G
(SS of 1460 bytes, some paths ha)-2.75 E .33 -.165(ve a s)-.22 H
(lightly).165 E(smaller MSS due to tunnels \(e.g., IPv6 o)117.3 398.4 Q
-.165(ve)-.165 G 2.75(rI).165 G 2.75(Pv4\). In)-2.75 F
(some speci\214c cases, IPv4 paths)2.75 E(may e)117.3 411.4 Q
(xperience a much smaller path MTU.)-.165 E
(Due to the complications of estimating the)5.5 E
(path MTU, and to the f)117.3 424.4 Q
(act that most paths support an MSS of at least 536 bytes, TFRC-)-.11 E
(SP as a def)117.3 437.4 Q(ault uses a nominal se)-.11 E
(gment size of 1460 bytes.)-.165 E 12.65<8354>100.8 454 S(aking pack)
-13.53 E(et headers into account: The allo)-.11 E
(wed transmit rate X in bytes per second is)-.275 E(reduced by a f)117.3
467 Q(actor that accounts for pack)-.11 E(et header size.)-.11 E
(This gi)5.5 E -.165(ve)-.275 G 2.75(st).165 G(he application)-2.75 E
(some incenti)117.3 480 Q -.165(ve)-.275 G 2.75(,b).165 G -.165(ey)-2.75
G(ond the Min Interv).165 E(al, not to use unnecessarily small pack)
-.275 E 2.75(ets. In)-.11 F(particular)117.3 493 Q 2.75(,w)-.44 G 2.75
(ei)-2.75 G(ntroduce a ne)-2.75 E 2.75(wp)-.275 G
(arameter H, which represents the e)-2.75 E(xpected size in bytes)-.165
E(of netw)117.3 506 Q
(ork and transport headers to be used on the TFRC connection')-.11 E
2.75(sp)-.605 G(ack)-2.75 E 2.75(ets. This)-.11 F(is)2.75 E
(used to reduce the allo)117.3 519 Q(wed transmit rate X as follo)-.275
E(ws:)-.275 E 2.75(X:)117.3 535.6 S 2.75(=X*s)-2.75 G
(_true / \(s_true + H\),)-2.75 E(where s_true is the true a)117.3 552.2
Q -.165(ve)-.22 G(rage data se).165 E
(gment size for the connection in bytes,)-.165 E -.165(ex)117.3 565.2 S
(cluding the transport and netw).165 E(ork headers.)-.11 E
(Section 4.1 of RFC 3448 states that)5.5 E(where the pack)117.3 578.2 Q
(et size v)-.11 E
(aries naturally with the data, an estimate of the mean se)-.275 E
(gment)-.165 E(size can be used for s_true.)117.3 591.2 Q
(As suggested in Section 4.1 of [RFC3448bis], when an)5.5 E
(estimate of the mean se)117.3 604.2 Q
(gment size is used for s_true, the estimate SHOULD be)-.165 E
(calculated o)117.3 617.2 Q -.165(ve)-.165 G 2.75(ra).165 G 2.75(tl)
-2.75 G(east the last four loss interv)-2.75 E 2.75(als. Ho)-.275 F(we)
-.275 E -.165(ve)-.275 G .88 -.44(r, t).165 H(his document does not).44
E(specify a speci\214c algorithm for estimating the mean se)117.3 630.2
Q(gment size.)-.165 E(The H parameter is set to the constant 40 bytes.)
117.3 646.8 Q(Thus, if the TFRC-SP application used)5.5 E
(40-byte data se)117.3 659.8 Q(gments, the allo)-.165 E
(wed transmit rate X w)-.275 E(ould be halv)-.11 E
(ed to account for the)-.165 E -.11(fa)117.3 672.8 S
(ct that half of the sending rate w).11 E(ould be used by the headers.)
-.11 E(Section 4.2 justi\214es this)5.5 E 2.75(de\214nition. Ho)117.3
685.8 R(we)-.275 E -.165(ve)-.275 G .88 -.44(r, a c).165 H
(onnection using TFRC-SP MA).44 E 2.75(Yi)-1.155 G
(nstead use a more precise)-2.75 E(Flo)72 733 Q(yd/K)-.11 E 284.171
(ohler Section)-.385 F 2.75(3. [P)2.75 F(age 8])-.165 E 0 Cg EP
%%Page: 9 9
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E
(estimate of H, based on the actual netw)117.3 85 Q
(ork and transport headers to be used on the)-.11 E(connection')117.3 98
Q 2.75(sp)-.605 G(ack)-2.75 E 2.75(ets. F)-.11 F(or e)-.165 E
(xample, a DCCP connection [RFC4340] o)-.165 E -.165(ve)-.165 G 2.75(rI)
.165 G(Pv4, where)-2.75 E(data pack)117.3 111 Q
(ets use the DCCP-Data pack)-.11 E
(et type, and there are no IP or DCCP options,)-.11 E
(could set H to 20 + 12 = 32 bytes.)117.3 124 Q
(If the TFRC implementation kno)5.5 E(ws that the IP layer)-.275 E
(is using IPv6 instead of IPv4, then the connection using TFRC-SP MA)
117.3 137 Q 2.75(Yu)-1.155 G(se an)-2.75 E(estimate of 40 bytes instead\
 of 60 bytes for H, for simplicity of implementation.)117.3 150 Q 12.65
<834d>100.8 166.6 S(easuring the loss e)-12.65 E -.165(ve)-.275 G
(nt rate in times of high loss: During short loss interv).165 E
(als \(those at)-.275 E(most tw)117.3 179.6 Q 2.75(or)-.11 G(ound-trip \
times\), the loss rate is computed by counting the actual number of)
-2.75 E(pack)117.3 192.6 Q(ets lost or mark)-.11 E
(ed, not by counting at most one loss e)-.11 E -.165(ve)-.275 G
(nt per loss interv).165 E(al.)-.275 E -.44(Wi)117.3 205.6 S
(thout this change, TFRC-SP could send multiple pack).44 E
(ets per round-trip time e)-.11 E -.165(ve)-.275 G 2.75(ni).165 G(n)
-2.75 E(the f)117.3 218.6 Q(ace of hea)-.11 E
(vy congestion, for a steady-state beha)-.22 E(vior of multiple pack)
-.22 E(ets dropped)-.11 E(each round-trip time.)117.3 231.6 Q
(In standard TFRC, the TFRC recei)117.3 248.2 Q -.165(ve)-.275 G 2.75
(re).165 G(stimates the loss e)-2.75 E -.165(ve)-.275 G
(nt rate by calculating the).165 E -2.475 -.22(av e)117.3 261.2 T
(rage loss interv).22 E(al in pack)-.275 E(ets, and in)-.11 E -.165(ve)
-.44 G(rting to get the loss e).165 E -.165(ve)-.275 G(nt rate.).165 E
(Thus, for a)5.5 E(short loss interv)117.3 274.2 Q(al with N pack)-.275
E(ets and K losses, standard TFRC calculates the size of)-.11 E
(that loss interv)117.3 287.2 Q(al as N pack)-.275 E(ets, contrib)-.11 E
(uting to a loss e)-.22 E -.165(ve)-.275 G(nt rate of 1/N.).165 E(Ho)5.5
E(we)-.275 E -.165(ve)-.275 G .88 -.44(r, f).165 H(or).44 E(TFRC-SP)
117.3 300.2 Q 2.75(,f)-1.221 G(or small loss interv)-2.75 E
(als of at most tw)-.275 E 2.75(or)-.11 G
(ound-trip times, if the loss interv)-2.75 E(al)-.275 E
(consists of N pack)117.3 313.2 Q
(ets including K losses, the size of the loss interv)-.11 E
(al is calculated as)-.275 E(N/K, contrib)117.3 326.2 Q
(uting to a loss e)-.22 E -.165(ve)-.275 G
(nt rate of K/N instead of 1/N.).165 E
(Section 5.4 of RFC 3448 speci\214es that the calculation of the a)117.3
342.8 Q -.165(ve)-.22 G(rage loss interv).165 E(al)-.275 E
(includes the most recent loss interv)117.3 355.8 Q
(al only if this increases the calculated a)-.275 E -.165(ve)-.22 G
(rage loss).165 E(interv)117.3 368.8 Q(al, as in the pseudocode belo)
-.275 E 4.18 -.715(w. H)-.275 H -.275(ow).715 G -2.365 -.275(ev e).275 H
.88 -.44(r, i).275 H 2.75(nT).44 G(FRC-SP the calculated loss interv)
-2.75 E(al)-.275 E(size for a short loss interv)117.3 381.8 Q(al v)-.275
E(aries as a function of the number of pack)-.275 E(et losses that)-.11
E(ha)117.3 394.8 Q .33 -.165(ve b)-.22 H(een detected, allo).165 E
(wing either increases or decreases in the calculated loss interv)-.275
E(al)-.275 E(size for the current short loss interv)117.3 407.8 Q
(al as ne)-.275 E 2.75(wp)-.275 G(ack)-2.75 E(ets are recei)-.11 E -.165
(ve)-.275 G 2.75(d. Therefore,).165 F(TFRC-)2.75 E
(SP adds the restriction that the calculation of the a)117.3 420.8 Q
-.165(ve)-.22 G(rage loss interv).165 E(al can include the)-.275 E
(most recent loss interv)117.3 433.8 Q(al only if more than tw)-.275 E
2.75(or)-.11 G(ound-trip times ha)-2.75 E .33 -.165(ve p)-.22 H
(assed since the).165 E(be)117.3 446.8 Q(ginning of that loss interv)
-.165 E(al.)-.275 E(Let the most recent loss interv)117.3 463.4 Q
(als be I_0 to I_n, with I_0 being the interv)-.275 E(al including the)
-.275 E(most recent loss e)117.3 476.4 Q -.165(ve)-.275 G
(nt, with the corresponding weights w_i as de\214ned in RFC 3448.).165 E
(In)5.5 E(RFC 3448 \(Section 5.4\), the a)117.3 489.4 Q -.165(ve)-.22 G
(rage loss interv).165 E(al I_mean is calculated as follo)-.275 E(ws:)
-.275 E(I_tot0 = 0;)169.8 508.4 Q(I_tot1 = 0;)169.8 521.4 Q(W_tot = 0;)
169.8 534.4 Q(for \(i = 0 to n-1\) {)169.8 547.4 Q
(I_tot0 = I_tot0 + \(I_i * w_i\);)175.3 560.4 Q(W_tot = W_tot + w_i;)
175.3 573.4 Q(})169.8 586.4 Q(for \(i = 1 to n\) {)169.8 599.4 Q
(I_tot1 = I_tot1 + \(I_i * w_\(i-1\)\);)175.3 612.4 Q(})169.8 625.4 Q
(I_tot = max\(I_tot0, I_tot1\);)169.8 638.4 Q(I_mean = I_tot/W_tot;)
169.8 651.4 Q(In TFRC-SP)117.3 674 Q 2.75(,t)-1.221 G(he a)-2.75 E -.165
(ve)-.22 G(rage loss interv).165 E
(al I_mean is instead calculated as follo)-.275 E(ws:)-.275 E(Flo)72 733
Q(yd/K)-.11 E 284.171(ohler Section)-.385 F 2.75(3. [P)2.75 F(age 9])
-.165 E 0 Cg EP
%%Page: 10 10
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E
(I_tot0 = 0;)169.8 85 Q(I_tot1 = 0;)169.8 98 Q(W_tot = 0;)169.8 111 Q
(for \(i = 0 to n-1\) {)169.8 124 Q(I_tot0 = I_tot0 + \(I_i * w_i\);)
175.3 137 Q(W_tot = W_tot + w_i;)175.3 150 Q(})169.8 163 Q
(for \(i = 1 to n\) {)169.8 176 Q
(I_tot1 = I_tot1 + \(I_i * w_\(i-1\)\);)175.3 189 Q(})169.8 202 Q
(If the current loss interv)169.8 215 Q(al I_0 is "short")-.275 E
(then I_tot = I_tot1;)175.3 228 Q(else I_tot = max\(I_tot0, I_tot1\);)
175.3 241 Q(I_mean = I_tot/W_tot;)169.8 254 Q 12.65<8341>100.8 276.6 S
(minimum interv)-9.9 E(al between pack)-.275 E
(ets: TFRC-SP enforces a Min Interv)-.11 E(al between)-.275 E(pack)117.3
289.6 Q(ets of 10 ms.)-.11 E 2.75<418d>5.5 G .55 -.275(ow t)-2.75 H
(hat wishes its transport protocol to e).275 E(xceed this Min Interv)
-.165 E(al)-.275 E(MUST use the con)117.3 302.6 Q -.165(ve)-.44 G
(ntional TFRC equations, rather than TFRC-SP).165 E 5.5(.T)-1.221 G
(he moti)-5.5 E -.275(va)-.275 G(tion).275 E(for this is discussed belo)
117.3 315.6 Q -.715(w.)-.275 G/F1 11/Times-Bold@0 SF(4.)72 341.6 Q/F2 14
/Times-Bold@0 SF(TFRC-SP Discussion)5.5 E F1(4.1.)72 367.6 Q/F3 13
/Times-Bold@0 SF(Response Functions and Thr)5.5 E(oughput Equations)
-.234 E F0(TFRC uses the TCP throughput equation gi)100.8 384.2 Q -.165
(ve)-.275 G 2.75(ni).165 G 2.75(n[)-2.75 G
(RFC3448], with the sending rate X in)-2.75 E(bytes per second as follo)
100.8 397.2 Q(ws:)-.275 E(s)208.3 416.2 Q 2.75(X=-)136.8 429.2 S
(------------------------------------------------------ ,)-2.75 E
(R*sqrt\(2*p/3\) + \(4*R* \(3*sqrt\(3*p/8\) * p * \(1+32*p^2\)\)\))147.8
442.2 Q(where:)100.8 461.2 Q/F4 11/Times-Italic@0 SF(s)129.6 477.8 Q F0
(is the pack)2.75 E(et size in bytes;)-.11 E F4(R)129.6 494.4 Q F0
(is the round trip time in seconds;)2.75 E F4(p)129.6 511 Q F0
(is the loss e)2.75 E -.165(ve)-.275 G
(nt rate, between 0 and 1.0, of the number of loss e).165 E -.165(ve)
-.275 G(nts as a fraction).165 E(of the number of pack)129.6 524 Q
(ets transmitted.)-.11 E(This equation uses an R)100.8 540.6 Q .396
-.198(TO o)-.66 H 2.75(f4).198 G
(*R, and assumes that the TCP connection sends an)-2.75 E(ackno)100.8
553.6 Q(wledgement for e)-.275 E -.165(ve)-.275 G(ry data pack).165 E
(et.)-.11 E(This equation essentially gi)100.8 570.2 Q -.165(ve)-.275 G
2.75(st).165 G
(he response function for TCP as well as for standard TFRC)-2.75 E
(\(modulo TCP')100.8 583.2 Q 2.75(sr)-.605 G
(ange of sender algorithms for setting the R)-2.75 E -.198(TO)-.66 G
2.75(\). As).198 F(sho)2.75 E(wn in T)-.275 E(able 1 of)-.88 E
([RFC3714], for high pack)100.8 596.2 Q
(et drop rates, this throughput equation gi)-.11 E -.165(ve)-.275 G 2.75
(sr).165 G(ough f)-2.75 E(airness with)-.11 E(the most aggressi)100.8
609.2 Q .33 -.165(ve p)-.275 H(ossible current TCP: a SA).165 E
(CK TCP \215o)-.44 E 2.75(wu)-.275 G(sing timestamps and ECN.)-2.75 E(B\
ecause it is not recommended for routers to use ECN-marking in highly-c\
ongested)100.8 622.2 Q(en)100.8 635.2 Q(vironments with high pack)-.44 E
(et dropping/marking rates [RFC3168] \(Section 7\), we note)-.11 E
(that it w)100.8 648.2 Q(ould be useful to ha)-.11 E .33 -.165(ve a t)
-.22 H(hroughput equation with a some).165 E(what more moderate)-.275 E
(sending rate for pack)100.8 661.2 Q(et drop rates of 40% and abo)-.11 E
-.165(ve)-.165 G(.).165 E(The ef)100.8 677.8 Q(fecti)-.275 E .33 -.165
(ve r)-.275 H(esponse function of TFRC-SP can be deri).165 E -.165(ve)
-.275 G 2.75(df).165 G(rom the TFRC response)-2.75 E
(function by using a se)100.8 690.8 Q
(gment size s of 1460 bytes, and using the loss e)-.165 E -.165(ve)-.275
G(nt rate actually).165 E(Flo)72 733 Q(yd/K)-.11 E 270.421
(ohler Section)-.385 F 2.75(4.1. [P)2.75 F(age 10])-.165 E 0 Cg EP
%%Page: 11 11
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E -.165
(ex)100.8 85 S(perienced by the TFRC-SP \215o).165 E 4.18 -.715(w. I)
-.275 H 2.75(na).715 G(ddition, for loss interv)-2.75 E
(als of at most tw)-.275 E 2.75(or)-.11 G(ound-trip)-2.75 E
(times, the loss e)100.8 98 Q -.165(ve)-.275 G
(nt rate for TFRC-SP is estimated by counting the actual number of lost)
.165 E(or mark)100.8 111 Q(ed pack)-.11 E
(ets, rather than by counting loss e)-.11 E -.165(ve)-.275 G 2.75
(nts. In).165 F(addition, the sending rate for)2.75 E
(TFRC-SP is constrained to be at most 100 pack)100.8 124 Q
(ets per second.)-.11 E -.165(Fo)100.8 140.6 S 2.75(ra).165 G 2.75(ne)
-2.75 G -.44(nv)-2.75 G(ironment with a \214x).44 E(ed pack)-.165 E
(et drop rate p, re)-.11 E -.055(ga)-.165 G(rdless of pack).055 E
(et size, the response)-.11 E(functions of TCP)100.8 153.6 Q 2.75(,T)
-1.221 G(FRC, and TFRC-SP are illustrated as follo)-2.75 E(ws, gi)-.275
E -.165(ve)-.275 G 2.75(ni).165 G 2.75(nK)-2.75 G(Bps \(Kilo)-2.75 E
(Bytes per second\), for a \215o)100.8 166.6 Q 2.75(ww)-.275 G
(ith a round-trip time of 100 ms:)-2.75 E/F1 11/Courier@0 SF 6.6
(<-- TCP)246 185.6 R(and Standard TFRC)6.6 E(-->)13.2 E 26.4
(Packet 14-byte)173.4 198.6 R 13.2(536-byte 1460-byte)26.4 F 13.2
(DropRate Segments Segments Segments)173.4 211.6 R 13.2
(-------- -------- -------- --------)173.4 224.6 R 39.6(0.00001 209.25)
173.4 237.6 R 19.8(2232.00 5812.49)26.4 F 39.6(0.00003 120.79)173.4
250.6 R 19.8(1288.41 3355.24)26.4 F 46.2(0.00010 66.12)173.4 263.6 R
19.8(705.25 1836.58)33 F 46.2(0.00030 38.10)173.4 276.6 R 19.8
(406.44 1058.45)33 F 46.2(0.00100 20.74)173.4 289.6 R 26.4
(221.23 576.12)33 F 46.2(0.00300 11.76)173.4 302.6 R 26.4(125.49 326.79)
33 F 52.8(0.01000 6.07)173.4 315.6 R 26.4(64.75 168.61)39.6 F 52.8
(0.03000 2.99)173.4 328.6 R 33(31.90 83.07)39.6 F 52.8(0.10000 0.96)
173.4 341.6 R 33(10.21 26.58)39.6 F 52.8(0.20000 0.29)173.4 354.6 R 39.6
(3.09 8.06)46.2 F 52.8(0.30000 0.11)173.4 367.6 R 39.6(1.12 2.93)46.2 F
52.8(0.40000 0.05)173.4 380.6 R 39.6(0.48 1.26)46.2 F 52.8(0.50000 0.02)
173.4 393.6 R 39.6(0.24 0.63)46.2 F
(Table 1: Response Function for TCP and TFRC.)166.8 419.6 Q
(Sending Rate in KBps, as a Function of Packet Drop Rate.)127.2 432.6 Q
F0(Flo)72 733 Q(yd/K)-.11 E 270.421(ohler Section)-.385 F 2.75(4.1. [P)
2.75 F(age 11])-.165 E 0 Cg EP
%%Page: 12 12
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E/F1 11
/Courier@0 SF 6.6(<---------- TFRC-SP -------->)246 85 R 26.4
(Packet 14-byte)173.4 98 R 13.2(536-byte 1460-byte)26.4 F 13.2
(DropRate Segments Segments Segments)173.4 111 R 13.2
(-------- -------- -------- --------)173.4 124 R 39.6(0.00001 5.40)173.4
137 R 26.4(57.60 150.00)39.6 F 39.6(0.00003 5.40)173.4 150 R 26.4
(57.60 150.00)39.6 F 39.6(0.00010 5.40)173.4 163 R 26.4(57.60 150.00)
39.6 F 39.6(0.00030 5.40)173.4 176 R 26.4(57.60 150.00)39.6 F 39.6
(0.00100 5.40)173.4 189 R 26.4(57.60 150.00)39.6 F 39.6(0.00300 5.40)
173.4 202 R 26.4(57.60 150.00)39.6 F 39.6(0.01000 5.40)173.4 215 R 26.4
(57.60 150.00)39.6 F 39.6(0.03000 5.40)173.4 228 R 33(57.60 83.07)39.6 F
39.6(0.10000 5.40)173.4 241 R 33(26.58 26.58)39.6 F 39.6
(0.20000 5.40 8.06 8.06)173.4 254 R 39.6(0.30000 2.93 2.93 2.93)173.4
267 R 39.6(0.40000 1.26 1.26 1.26)173.4 280 R 39.6
(0.50000 0.63 0.63 0.63)173.4 293 R
(Table 2: Response Function for TFRC-SP.)180 319 Q
(Sending Rate in KBps, as a Function of Packet Drop Rate.)127.2 332 Q
(Maximum Sending Rate of 100 Packets per Second.)153.6 345 Q F0
(The calculations for T)100.8 367.6 Q(ables 1 and 2 use the pack)-.88 E
(et loss rate for an approximation for the)-.11 E(loss e)100.8 380.6 Q
-.165(ve)-.275 G(nt rate p.).165 E
(Scripts and graphs for the tables are a)5.5 E -.275(va)-.22 G
(ilable from [V).275 E 2.75(OIPSIMS]. As)-.44 F(the)2.75 E(well-kno)
100.8 393.6 Q(wn TCP response function in T)-.275 E(able 1 sho)-.88 E
(ws, the sending rate for TCP and)-.275 E(standard TFRC v)100.8 406.6 Q
(aries linearly with se)-.275 E(gment size.)-.165 E
(The TFRC-SP response function sho)5.5 E(wn)-.275 E(in T)100.8 419.6 Q
(able 2 re\215ects the maximum sending rate of a hundred pack)-.88 E
(ets per second; when not)-.11 E
(limited by this maximum sending rate, the TFRC-SP \215o)100.8 432.6 Q
2.75(wr)-.275 G(ecei)-2.75 E -.165(ve)-.275 G 2.75(st).165 G
(he same sending rate in)-2.75 E(KBps as the TCP \215o)100.8 445.6 Q
2.75(ww)-.275 G(ith 1460-byte se)-2.75 E(gments, gi)-.165 E -.165(ve)
-.275 G 2.75(nt).165 G(he same pack)-2.75 E(et drop rate.)-.11 E
(Simulations sho)100.8 458.6 Q(wing the TCP)-.275 E 2.75(,s)-1.221 G
(tandard TFRC, and TFRC-SP sending rates in response to a)-2.75 E
(con\214gured pack)100.8 471.6 Q(et drop rate are gi)-.11 E -.165(ve)
-.275 G 2.75(ni).165 G 2.75(nT)-2.75 G
(ables 7, 8, and 9, and are consistent with the)-3.63 E
(response functions sho)100.8 484.6 Q(wn here.)-.275 E(Flo)72 733 Q
(yd/K)-.11 E 270.421(ohler Section)-.385 F 2.75(4.1. [P)2.75 F(age 12])
-.165 E 0 Cg EP
%%Page: 13 13
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E/F1 11
/Courier@0 SF 6.6(<-- TCP)259.2 85 R(and Standard TFRC)6.6 E(-->)13.2 E
52.8(Byte 14-byte)173.4 98 R 13.2(536-byte 1460-byte)26.4 F 26.4
(DropRate Segments)173.4 111 R 13.2(Segments Segments)19.8 F 26.4
(-------- --------)173.4 124 R 13.2(-------- --------)19.8 F 26.4
(0.0000001 284.76 929.61)173.4 137 R(1498.95)26.4 E 26.4
(0.0000003 164.39 536.17 863.16)173.4 150 R 33(0.0000010 90.01)173.4 163
R 26.4(292.64 468.49)33 F 33(0.0000030 51.92)173.4 176 R 26.4
(167.28 263.68)33 F 33(0.0000100 28.34 88.56)173.4 189 R(132.75)33 E 33
(0.0000300 16.21 46.67 61.70)173.4 202 R 39.6(0.0001000 8.60)173.4 215 R
33(19.20 16.25)39.6 F 39.6(0.0003000 4.56 4.95 1.70)173.4 228 R 39.6
(0.0010000 1.90 0.37 0.15)173.4 241 R 39.6(0.0030000 0.52 0.05 0.06)
173.4 254 R 39.6(0.0100000 0.04 0.02 0.06)173.4 267 R 39.6
(0.0300000 0.00 0.02 0.06)173.4 280 R
(Table 3: Response Function for TCP and TFRC.)173.4 306 Q
(Sending Rate in KBps, as a Function of Byte Drop Rate.)133.8 319 Q 6.6
(<---------- TFRC-SP -------->)259.2 344 R 52.8(Byte 14-byte)173.4 357 R
13.2(536-byte 1460-byte)26.4 F 26.4(DropRate Segments)173.4 370 R 13.2
(Segments Segments)19.8 F 26.4(-------- --------)173.4 383 R 13.2
(-------- --------)19.8 F 39.6(0.0000001 5.40)173.4 396 R 26.4
(57.60 150.00)39.6 F 39.6(0.0000003 5.40)173.4 409 R 26.4(57.60 150.00)
39.6 F 39.6(0.0000010 5.40)173.4 422 R 26.4(57.60 150.00)39.6 F 39.6
(0.0000030 5.40)173.4 435 R 26.4(57.60 150.00)39.6 F 39.6
(0.0000100 5.40)173.4 448 R 26.4(57.60 132.75)39.6 F 39.6
(0.0000300 5.40)173.4 461 R 33(57.60 61.70)39.6 F 39.6(0.0001000 5.40)
173.4 474 R 33(50.00 16.25)39.6 F 39.6(0.0003000 5.40)173.4 487 R 39.6
(12.89 1.70)39.6 F 39.6(0.0010000 5.40 0.95 0.15)173.4 500 R 39.6
(0.0030000 5.40 0.12 0.06)173.4 513 R 39.6(0.0100000 1.10 0.06 0.06)
173.4 526 R 39.6(0.0300000 0.13 0.06 0.06)173.4 539 R
(Table 4: Response Function for TFRC-SP.)186.6 565 Q
(Sending Rate in KBps, as a Function of Byte Drop Rate.)133.8 578 Q
(Maximum Sending Rate of 100 Packets per Second.)153.6 591 Q F0 -.165
(Fo)100.8 613.6 S 2.75(rT).165 G(ables 3 and 4, the pack)-3.63 E
(et drop rate is calculated as 1-\(1-b\)^N, for a byte drop rate of b,)
-.11 E(and a pack)100.8 626.6 Q(et size of N bytes.)-.11 E
(These tables use the pack)5.5 E(et loss rate as an approximation for)
-.11 E(the loss e)100.8 639.6 Q -.165(ve)-.275 G(nt rate p.).165 E
(The TCP response functions sho)5.5 E(wn in T)-.275 E(able 3 for \214x)
-.88 E(ed byte drop)-.165 E(rates are rather dif)100.8 652.6 Q
(ferent from the response functions sho)-.275 E(wn in T)-.275 E
(able 1 for \214x)-.88 E(ed pack)-.165 E(et drop)-.11 E
(rates; with higher byte drop rates, a TCP connection can ha)100.8 665.6
Q .33 -.165(ve a h)-.22 H(igher sending rate using).165 E
(*smaller* pack)100.8 678.6 Q 2.75(ets. T)-.11 F(able 4 also sho)-.88 E
(ws that with \214x)-.275 E(ed byte drop rates, the sending rate for)
-.165 E(TFRC-SP can be signi\214cantly higher than that for TCP or stan\
dard TFRC, re)100.8 691.6 Q -.055(ga)-.165 G(rdless of the).055 E(Flo)72
733 Q(yd/K)-.11 E 270.421(ohler Section)-.385 F 2.75(4.1. [P)2.75 F
(age 13])-.165 E 0 Cg EP
%%Page: 14 14
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E(TCP se)
100.8 85 Q(gment size.)-.165 E(This is because in this en)5.5 E
(vironment, with small pack)-.44 E(ets, TFRC-SP)-.11 E(recei)100.8 98 Q
-.165(ve)-.275 G 2.75(sas).165 G(mall pack)-2.75 E(et drop rate, b)-.11
E(ut is allo)-.22 E(wed to send at the sending rate of a TCP or)-.275 E
(standard TFRC \215o)100.8 111 Q 2.75(wu)-.275 G(sing lar)-2.75 E
(ger pack)-.198 E(ets b)-.11 E(ut recei)-.22 E(ving the same pack)-.275
E(et drop rate.)-.11 E(Simulations sho)100.8 127.6 Q(wing TCP)-.275 E
2.75(,s)-1.221 G
(tandard TFRC, and TFRC-SP sending rates in response to a)-2.75 E
(con\214gured byte drop rate are gi)100.8 140.6 Q -.165(ve)-.275 G 2.75
(ni).165 G 2.75(nA)-2.75 G(ppendix B.2 .)-2.75 E/F1 11/Times-Bold@0 SF
(4.2.)72 166.6 Q/F2 13/Times-Bold@0 SF(Accounting f)5.5 E
(or Header Size)-.325 E F0([RFC3714] mak)100.8 183.2 Q
(es the optimistic assumption that the limitation of the netw)-.11 E
(ork is in)-.11 E
(bandwidth in bytes per second \(Bps\), and not in CPU c)100.8 196.2 Q
(ycles or in pack)-.165 E(ets per second \(pps\).)-.11 E(Ho)100.8 209.2
Q(we)-.275 E -.165(ve)-.275 G .88 -.44(r, s).165 H(ome attention must b\
e paid to the load in pps as well as to the load in Bps.).44 E(Ev)100.8
222.2 Q(en aside from the Min Interv)-.165 E(al, TFRC-SP gi)-.275 E
-.165(ve)-.275 G 2.75(st).165 G(he application some incenti)-2.75 E .33
-.165(ve t)-.275 H 2.75(ou).165 G(se)-2.75 E(fe)100.8 235.2 Q(wer b)
-.275 E(ut lar)-.22 E(ger pack)-.198 E(ets, when lar)-.11 E(ger pack)
-.198 E(ets w)-.11 E(ould suf)-.11 E(\214ce, by including the bandwidth)
-.275 E(used by the pack)100.8 248.2 Q(et header in the allo)-.11 E
(wed sending rate.)-.275 E(As an e)100.8 264.8 Q
(xample, a sender using 120-byte pack)-.165 E
(ets needs a TCP-friendly rate of 128 Kbps to)-.11 E
(send 96 Kbps of application data.)100.8 277.8 Q
(This is because the TCP-friendly rate is reduced by a)5.5 E -.11(fa)
100.8 290.8 S
(ctor of s_true/\(s_true + H\) = 120/160, to account for the ef).11 E
(fect of pack)-.275 E(et headers.)-.11 E(If the)5.5 E
(sender suddenly switched to 40-byte data se)100.8 303.8 Q
(gments, the allo)-.165 E(wed sending rate w)-.275 E(ould reduce)-.11 E
(to 64 Kbps of application data; and the use of one-byte data se)100.8
316.8 Q(gments w)-.165 E(ould reduce the)-.11 E(allo)100.8 329.8 Q
(wed sending rate to 3.12 Kbps of application data.)-.275 E(\(In f)5.5 E
(act, the Min Interv)-.11 E(al w)-.275 E(ould)-.11 E(pre)100.8 342.8 Q
-.165(ve)-.275 G(nt senders from achie).165 E
(ving these rates, since applications using TFRC-SP cannot send)-.275 E
(more than 100 pack)100.8 355.8 Q(ets per second.\))-.11 E(Unless it ha\
s a more precise estimate of the header size, TFRC-SP assumes 40 bytes \
for the)100.8 372.4 Q(header size, although the header could be lar)
100.8 385.4 Q(ger \(due to IP options, IPv6, IP tunnels, and)-.198 E
(the lik)100.8 398.4 Q
(e\) or smaller \(due to header compression\) on the wire.)-.11 E
(Requiring the use of the e)5.5 E(xact)-.165 E(header size in bytes w)
100.8 411.4 Q(ould require signi\214cant additional comple)-.11 E(xity)
-.165 E 2.75(,a)-.715 G(nd w)-2.75 E(ould ha)-.11 E .33 -.165(ve l)-.22
H(ittle).165 E(additional bene\214t.)100.8 424.4 Q(TFRC-SP')5.5 E 2.75
(sd)-.605 G(ef)-2.75 E(ault assumption of a 40-byte header is suf)-.11 E
(\214cient to get a)-.275 E(rough estimate of the throughput, and to gi)
100.8 437.4 Q .33 -.165(ve t)-.275 H(he application some incenti).165 E
.33 -.165(ve n)-.275 H(ot to use).165 E(unnecessarily-man)100.8 450.4 Q
2.75(ys)-.165 G(mall pack)-2.75 E 2.75(ets. Because)-.11 F
(we are only aiming at rough f)2.75 E(airness, and at a)-.11 E
(rough incenti)100.8 463.4 Q .33 -.165(ve f)-.275 H
(or applications, the def).165 E
(ault use of a 40-byte header in the calculations of)-.11 E
(the header bandwidth is suf)100.8 476.4 Q
(\214cient for both IPv4 and IPv6.)-.275 E F1(4.3.)72 502.4 Q F2
(The TFRC-SP Min Inter)5.5 E -.13(va)-.13 G(l).13 E F0
(The header size calculation pro)100.8 519 Q(vides an incenti)-.165 E
.33 -.165(ve f)-.275 H(or applications to use fe).165 E(wer)-.275 E 2.75
(,b)-.44 G(ut lar)-2.97 E(ger)-.198 E(,)-.44 E(pack)100.8 532 Q 2.75
(ets. Another)-.11 F(incenti)2.75 E .33 -.165(ve i)-.275 H 2.75(st).165
G(hat when the path limitation is in pps, the application using)-2.75 E
(more small pack)100.8 545 Q(ets is lik)-.11 E(ely to cause higher pack)
-.11 E(et drop rates, and to ha)-.11 E .33 -.165(ve t)-.22 H 2.75(or)
.165 G(educe its)-2.75 E(sending rate accordingly)100.8 558 Q 5.5(.T)
-.715 G(hat is, if the congestion is in terms of pps, then the \215o)
-5.5 E 2.75(ws)-.275 G(ending)-2.75 E(more pps will increase the pack)
100.8 571 Q(et drop rate, and ha)-.11 E .33 -.165(ve t)-.22 H 2.75(oa)
.165 G(djust its sending rate accordingly)-2.75 E(.)-.715 E(Ho)100.8 584
Q(we)-.275 E -.165(ve)-.275 G .88 -.44(r, t).165 H
(he increased congestion caused by the use of small pack).44 E
(ets in an en)-.11 E(vironment)-.44 E(limited by pps is e)100.8 597 Q
(xperienced not only by the \215o)-.165 E 2.75(wu)-.275 G
(sing the small pack)-2.75 E(ets, b)-.11 E(ut by all of the)-.22 E
(competing traf)100.8 610 Q(\214c on that congested link.)-.275 E
(These incenti)5.5 E -.165(ve)-.275 G 2.75(sa).165 G(re therefore insuf)
-2.75 E(\214cient to)-.275 E(pro)100.8 623 Q(vide suf)-.165 E
(\214cient protection for pps netw)-.275 E(ork limitations.)-.11 E
(TFRC-SP)100.8 639.6 Q 2.75(,t)-1.221 G(hen, includes a Min Interv)-2.75
E(al of 10 ms.)-.275 E(This pro)5.5 E(vides additional restrictions on)
-.165 E(the use of unnecessarily man)100.8 652.6 Q 2.75(ys)-.165 G
(mall pack)-2.75 E(ets.)-.11 E(One justi\214cation for the Min Interv)
100.8 669.2 Q
(al is the practical one that the applications that currently)-.275 E
-.11(wa)100.8 682.2 S(nt to send small pack).11 E(ets are the V)-.11 E
(oIP applications that send at most one pack)-1.419 E(et e)-.11 E -.165
(ve)-.275 G(ry 10).165 E(ms, so this restriction does not af)100.8 695.2
Q(fect current traf)-.275 E 2.75(\214c. A)-.275 F
(second justi\214cation is that there is)2.75 E(Flo)72 733 Q(yd/K)-.11 E
270.421(ohler Section)-.385 F 2.75(4.3. [P)2.75 F(age 14])-.165 E 0 Cg
EP
%%Page: 15 15
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E
(no pressing need for best-ef)100.8 85 Q(fort traf)-.275 E
(\214c in the current Internet to send small pack)-.275 E(ets more)-.11
E(frequently than once e)100.8 98 Q -.165(ve)-.275 G
(ry 10 ms \(rather than taking the 10 ms delay at the sender).165 E 2.75
(,a)-.44 G(nd)-2.75 E(mer)100.8 111 Q(ging the tw)-.198 E 2.75(os)-.11 G
(mall pack)-2.75 E(ets into one lar)-.11 E(ger one\).)-.198 E
(This 10 ms delay for mer)5.5 E(ging small)-.198 E(pack)100.8 124 Q
(ets is lik)-.11 E(ely to be dominated by the netw)-.11 E(ork propag)
-.11 E(ation, transmission, and queueing)-.055 E(delays of best-ef)100.8
137 Q(fort traf)-.275 E(\214c in the current Internet.)-.275 E
(As a result, our judgment w)5.5 E(ould be that)-.11 E
(the bene\214t to the user of ha)100.8 150 Q
(ving less than 10 ms between pack)-.22 E(ets is outweighed by the)-.11
E(bene\214t to the netw)100.8 163 Q(ork of a)-.11 E -.22(vo)-.22 G
(iding unnecessarily man).22 E 2.75(ys)-.165 G(mall pack)-2.75 E(ets.)
-.11 E(The Min Interv)100.8 179.6 Q
(al causes TFRC-SP not to support applications sending small pack)-.275
E(ets v)-.11 E(ery)-.165 E(frequently)100.8 192.6 Q 5.5(.C)-.715 G
(onsider a TFRC \215o)-5.5 E 2.75(ww)-.275 G(ith a \214x)-2.75 E
(ed pack)-.165 E(et size of 100 bytes, b)-.11 E(ut with a v)-.22 E
(ariable)-.275 E(sending rate and a f)100.8 205.6 Q
(airly uncongested path.)-.11 E(When this \215o)5.5 E 2.75(ww)-.275 G
(as sending at most 100 pps, it)-2.86 E -.11(wo)100.8 218.6 S
(uld be able to use TFRC-SP).11 E 5.5(.I)-1.221 G 2.75(ft)-5.5 G
(he \215o)-2.75 E 2.75(ww)-.275 G
(ished to increase its sending rate to more than)-2.75 E(100 pps, b)
100.8 231.6 Q(ut to k)-.22 E(eep the same pack)-.11 E(et size, it w)-.11
E(ould no longer be able to achie)-.11 E .33 -.165(ve t)-.275 H
(his with).165 E(TFRC-SP)100.8 244.6 Q 2.75(,a)-1.221 G(nd w)-2.75 E
(ould ha)-.11 E .33 -.165(ve t)-.22 H 2.75(os).165 G(witch to the def)
-2.75 E(ault TFRC, recei)-.11 E(ving a dramatic,)-.275 E
(discontinuous decrease in its allo)100.8 257.6 Q(wed sending rate.)
-.275 E(This seems not only acceptable, b)5.5 E(ut)-.22 E
(desirable for the global Internet.)100.8 270.6 Q(What is to pre)100.8
287.2 Q -.165(ve)-.275 G(nt \215o).165 E
(ws from opening multiple connections, each with a 10 ms Min)-.275 E
(Interv)100.8 300.2 Q
(al, and thereby getting around the limitation of the Min Interv)-.275 E
2.75(al? Ob)-.275 F(viously)-.165 E 2.75(,t)-.715 G(here is)-2.75 E
(nothing to pre)100.8 313.2 Q -.165(ve)-.275 G(nt \215o).165 E
(ws from doing this, just as there is currently nothing to pre)-.275 E
-.165(ve)-.275 G(nt \215o).165 E(ws)-.275 E(from using UDP)100.8 326.2 Q
2.75(,o)-1.221 G 2.75(rf)-2.75 G
(rom opening multiple parallel TCP connections, or from using their)
-2.75 E -.275(ow)100.8 339.2 S 2.75(nc).275 G
(ongestion control mechanism.)-2.75 E
(Of course, implementations or middlebox)5.5 E(es are also)-.165 E(free\
 to limit the number of parallel TFRC connections opened to the same de\
stination in)100.8 352.2 Q
(times of congestion, if that seems desirable.)100.8 365.2 Q(And \215o)
5.5 E(ws that open multiple parallel)-.275 E
(connections are subject to the incon)100.8 378.2 Q -.165(ve)-.44 G
(niences of reordering and the lik).165 E(e.)-.11 E/F1 11/Times-Bold@0
SF(4.4.)72 404.2 Q/F2 13/Times-Bold@0 SF(Counting P)5.5 E(ack)-.13 E
(et Losses)-.13 E F0(It is not possible for a TCP connection to persist\
ently send multiple pack)100.8 420.8 Q(ets per round-trip)-.11 E
(time in the f)100.8 433.8 Q
(ace of high congestion, with a steady-state with multiple pack)-.11 E
(ets dropped per)-.11 E(round-trip time.)100.8 446.8 Q -.165(Fo)5.5 G
2.75(rT).165 G(CP)-2.75 E 2.75(,w)-1.221 G(hen one or more pack)-2.75 E
(ets are dropped each round-trip time, the)-.11 E
(sending rate is quickly dropped to less than one pack)100.8 459.8 Q
(et per round-trip time.)-.11 E(In addition, for)5.5 E(TCP with T)100.8
472.8 Q(ahoe, Ne)-.88 E(wReno, or SA)-.275 E
(CK congestion control mechanisms, the response to)-.44 E
(congestion is lar)100.8 485.8 Q(gely independent of the number of pack)
-.198 E(ets dropped per round-trip time.)-.11 E
(As a result, standard TFRC can best achie)100.8 502.4 Q .33 -.165(ve f)
-.275 H(airness with TCP).055 E 2.75(,e)-1.221 G -.165(ve)-3.025 G 2.75
(ni).165 G 2.75(nh)-2.75 G(ighly congested)-2.75 E(en)100.8 515.4 Q
(vironments, by calculating the loss e)-.44 E -.165(ve)-.275 G
(nt rate rather than the pack).165 E(et drop rate, where a)-.11 E
(loss e)100.8 528.4 Q -.165(ve)-.275 G(nt is one or more pack).165 E
(ets dropped or mark)-.11 E(ed from a windo)-.11 E 2.75(wo)-.275 G 2.75
(fd)-2.75 G(ata.)-2.75 E(Ho)100.8 545 Q(we)-.275 E -.165(ve)-.275 G .88
-.44(r, w).165 H(ith TFRC-SP).44 E 2.75(,i)-1.221 G 2.75(ti)-2.75 G 2.75
(sn)-2.75 G 2.75(ol)-2.75 G(onger possible to achie)-2.75 E .33 -.165
(ve f)-.275 H(airness with TCP or with).055 E
(standard TFRC by counting only the loss e)100.8 558 Q -.165(ve)-.275 G
(nt rate [WBL04].).165 E(Instead of sending one)5.5 E(lar)100.8 571 Q
(ge pack)-.198 E
(et per round-trip time, TFRC-SP could be sending N small pack)-.11 E
(ets \(where N)-.11 E(small pack)100.8 584 Q(ets equal one lar)-.11 E
(ge 1500-byte pack)-.198 E 2.75(et\). The)-.11 F
(loss measurement used with TFRC-)2.75 E
(SP has to be able to detect a connection that is consistently recei)
100.8 597 Q(ving multiple pack)-.275 E(et losses)-.11 E
(or marks per round-trip time, to allo)100.8 610 Q 2.75(wT)-.275 G
(FRC-SP to respond appropriately)-2.75 E(.)-.715 E(In TFRC-SP)100.8
626.6 Q 2.75(,t)-1.221 G(he loss e)-2.75 E -.165(ve)-.275 G
(nt rate is calculated by counting at most one loss e).165 E -.165(ve)
-.275 G(nt in loss).165 E(interv)100.8 639.6 Q(als longer than tw)-.275
E 2.75(or)-.11 G(ound-trip times, and by counting each pack)-2.75 E
(et lost or mark)-.11 E(ed in)-.11 E(shorter loss interv)100.8 652.6 Q
2.75(als. In)-.275 F(particular)2.75 E 2.75(,f)-.44 G
(or a short loss interv)-2.75 E(al with N pack)-.275 E(ets, including K)
-.11 E(lost or mark)100.8 665.6 Q(ed pack)-.11 E(ets, the loss interv)
-.11 E(al length is calculated as N/K, instead as N.)-.275 E(The)5.5 E
-2.475 -.22(av e)100.8 678.6 T(rage loss interv).22 E
(al I_mean is still a)-.275 E -.165(ve)-.22 G(raged o).165 E -.165(ve)
-.165 G 2.75(rt).165 G(he most recent eight loss interv)-2.75 E(als, as)
-.275 E(speci\214ed in Section 5.4 of RFC 3448.)100.8 691.6 Q
(Thus, if eight successi)5.5 E .33 -.165(ve l)-.275 H(oss interv).165 E
(als are short loss)-.275 E(Flo)72 733 Q(yd/K)-.11 E 270.421
(ohler Section)-.385 F 2.75(4.4. [P)2.75 F(age 15])-.165 E 0 Cg EP
%%Page: 16 16
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E(interv)
100.8 85 Q(als with N pack)-.275 E(ets and K losses, the loss e)-.11 E
-.165(ve)-.275 G(nt rate is calculated as K/N, rather than).165 E
(as 1/N.)100.8 98 Q/F1 11/Times-Bold@0 SF(4.5.)72 124 Q/F2 13
/Times-Bold@0 SF(The Nominal P)5.5 E(ack)-.13 E(et Size)-.13 E F1 2.75
(4.5.1. P)72 150 R(ack)-.11 E(et Size and P)-.11 E(ack)-.11 E(et Dr)-.11
E(op Rates)-.198 E F0(The guidelines in Section 3 abo)100.8 166.6 Q .33
-.165(ve s)-.165 H(ay that the nominal se).165 E
(gment size s is set to 1460 bytes,)-.165 E(pro)100.8 179.6 Q
(viding a goal of f)-.165 E
(airness, in terms of the sending rate in bytes per second, with a TCP)
-.11 E<8d6f>100.8 192.6 Q 2.75(ww)-.275 G
(ith 1460 bytes of application data per pack)-2.75 E(et b)-.11 E
(ut with the same pack)-.22 E(et drop rate.)-.11 E(This follo)100.8
205.6 Q(ws the assumption that a TCP \215o)-.275 E 2.75(ww)-.275 G
(ith 1460-byte se)-2.75 E(gments will ha)-.165 E .33 -.165(ve a h)-.22 H
(igher).165 E(sending rate than a TCP \215o)100.8 218.6 Q 2.75(ww)-.275
G(ith smaller se)-2.75 E 2.75(gments. While)-.165 F
(this assumption holds in an)2.75 E(en)100.8 231.6 Q
(vironment where the pack)-.44 E(et drop rate is independent of pack)
-.11 E(et size, this assumption does)-.11 E
(not necessarily hold in an en)100.8 244.6 Q(vironment where lar)-.44 E
(ger pack)-.198 E(ets are more lik)-.11 E(ely to be dropped)-.11 E
(than are small pack)100.8 257.6 Q(ets.)-.11 E(The table belo)100.8
274.2 Q 2.75(ws)-.275 G(ho)-2.75 E
(ws the results of simulations with standard \(SA)-.275 E
(CK\) TCP \215o)-.44 E(ws, where,)-.275 E(for each *byte*, the pack)
100.8 287.2 Q(et containing that byte is dropped with probability p.)
-.11 E(Consider)5.5 E
(the approximation for the TCP response function for pack)100.8 300.2 Q
(et drop rates up to 10% or so; for)-.11 E(this en)100.8 313.2 Q
(vironments, the sending rate in bytes per R)-.44 E
(TT is roughly 1.2 s/sqrt\(p\), for a pack)-.66 E(et)-.11 E
(size of s bytes and pack)100.8 326.2 Q(et drop rate p.)-.11 E(Gi)5.5 E
-.165(ve)-.275 G 2.75(na\214).165 G -.165(xe)-2.75 G 2.75(d*).165 G
(byte* drop rate p1, and a TCP pack)-2.75 E(et)-.11 E
(size of s bytes, the pack)100.8 339.2 Q
(et drop rate is roughly s*p1, producing a sending rate in bytes per)
-.11 E -.66(RT)100.8 352.2 S 2.75(To).66 G 2.75(fr)-2.75 G
(oughly 1.2 sqrt\(s\)/sqrt\(p1\).)-2.75 E(Thus, for TCP in an en)5.5 E
(vironment with a \214x)-.44 E(ed byte)-.165 E
(drop rate, the sending rate should gro)100.8 365.2 Q 2.75(wr)-.275 G
(oughly as sqrt\(s\), for pack)-2.75 E(et drop rates up to 10%)-.11 E
(or so.)100.8 378.2 Q(Each ro)100.8 394.8 Q 2.75(wo)-.275 G 2.75(fT)
-2.75 G(able 5 belo)-3.63 E 2.75(ws)-.275 G(ho)-2.75 E
(ws a separate simulation with ten TCP connection and a)-.275 E<8c78>
100.8 407.8 Q
(ed byte drop rate of 0.0001, with each simulation using a dif)-.165 E
(ferent se)-.275 E(gment size.)-.165 E -.165(Fo)5.5 G(r).165 E
(each simulation, the TCP sending rate and goodput are a)100.8 420.8 Q
-.165(ve)-.22 G(raged o).165 E -.165(ve)-.165 G 2.75(rt).165 G
(he ten \215o)-2.75 E 2.75(ws. As)-.275 F(one)2.75 E -.11(wo)100.8 433.8
S(uld e).11 E(xpect from the paragraph abo)-.165 E -.165(ve)-.165 G 2.75
(,t).165 G(he TCP sending rate gro)-2.75 E(ws some)-.275 E
(what less than)-.275 E(linearly with an increase in pack)100.8 446.8 Q
(et size, up to a pack)-.11 E(et size of 1460 bytes, corresponding to)
-.11 E 2.75(ap)100.8 459.8 S(ack)-2.75 E(et drop rate of 13%.)-.11 E
(After that, further increases in the pack)5.5 E(et size result in a)
-.11 E(*decrease* in the TCP sending rate, as the TCP connection enters\
 the re)100.8 472.8 Q(gime of)-.165 E -.165(ex)100.8 485.8 S
(ponential back).165 E(of)-.11 E 2.75(fo)-.275 G 2.75(ft)-2.75 G
(he retransmit timer)-2.75 E(.)-.605 E/F3 11/Courier@0 SF 13.2
(Segment Packet)193.2 504.8 R(TCP Rates \(Kbps\))39.6 E(Size \(B\))193.2
517.8 Q 13.2(DropRate SendRate)13.2 F(Goodput)26.4 E 6.6
(-------- --------)193.2 530.8 R 19.8(-------- -------)19.8 F 33
(14 0.005)219.6 543.8 R 39.6(6.37 6.34)46.2 F 33(128 0.016 30.78 30.30)
213 556.8 R 33(256 0.028 46.54 44.96)213 569.8 R 33
(512 0.053 62.43 58.37)213 582.8 R 33(1460 0.134 94.15 80.02)206.4 595.8
R 33(4000 0.324 35.20 21.44)206.4 608.8 R 33(8000 0.531 15.36)206.4
621.8 R(5.76)46.2 E(Table 5: TCP Median Send Rate vs. Packet Size I:)
173.4 647.8 Q(Byte Drop Rate 0.0001)259.2 660.8 Q F0 -.88(Ta)100.8 679.8
S(ble 6 belo).88 E 2.75(ws)-.275 G(ho)-2.75 E
(ws similar results for a byte drop rate of 0.001.)-.275 E
(In this case, the TCP)5.5 E(sending rate gro)100.8 692.8 Q
(ws with increasing pack)-.275 E(et size up to a pack)-.11 E
(et size of 128 bytes,)-.11 E(Flo)72 733 Q(yd/K)-.11 E 262.171
(ohler Section)-.385 F 2.75(4.5.1. [P)2.75 F(age 16])-.165 E 0 Cg EP
%%Page: 17 17
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E
(corresponding to a pack)100.8 85 Q(et drop rate of 16%.)-.11 E
(After than, the TCP sending rate decreases)5.5 E(and then increases ag)
100.8 98 Q(ain, as the TCP connection enters the re)-.055 E(gime of e)
-.165 E(xponential back)-.165 E(of)-.11 E(f)-.275 E
(of the retransmit timer)100.8 111 Q 5.5(.N)-.605 G
(ote that with this byte drop rate, with pack)-5.5 E
(et sizes of 4000 and)-.11 E
(8000 bytes, the TCP sending rate increases b)100.8 124 Q
(ut the TCP goodput rate remains essentially)-.22 E 2.75(zero. This)
100.8 137 R(mak)2.75 E(es sense, as almost all pack)-.11 E
(ets that are sent are dropped.)-.11 E/F1 11/Courier@0 SF 13.2
(Segment Packet)193.2 156 R(TCP Rates \(Kbps\))39.6 E(Size \(B\))193.2
169 Q 13.2(DropRate SendRate)13.2 F(Goodput)26.4 E 6.6
(-------- --------)193.2 182 R 19.8(-------- -------)19.8 F 33(14 0.053)
219.6 195 R 39.6(1.68 1.56)46.2 F 33(128 0.159)213 208 R 39.6(7.66 6.13)
46.2 F 33(256 0.248)213 221 R 39.6(6.21 4.32)46.2 F 33(512 0.402)213 234
R 39.6(1.84 1.11)46.2 F 33(1460 0.712)206.4 247 R 39.6(1.87 0.47)46.2 F
33(4000 0.870)206.4 260 R 39.6(3.20 0.00)46.2 F 33(8000 0.890)206.4 273
R 39.6(5.76 0.00)46.2 F
(Table 6: TCP Median Send Rate vs. Packet Size II:)173.4 299 Q
(Byte Drop Rate 0.001)259.2 312 Q F0(The TCP beha)100.8 331 Q
(vior in the presence of a \214x)-.22 E
(ed byte drop rate suggests that instead of the goal)-.165 E
(of a TFRC-SP \215o)100.8 344 Q 2.75(wa)-.275 G(chie)-2.75 E
(ving the same sending rate in bytes per second as a TCP \215o)-.275 E
(w)-.275 E(using 1500-byte pack)100.8 357 Q(ets, it mak)-.11 E
(es more sense to consider an ideal goal of a TFRC-SP \215o)-.11 E(w)
-.275 E(achie)100.8 370 Q(ving the same sending rate as a TCP \215o)
-.275 E 2.75(ww)-.275 G(ith the optimal pack)-2.75 E(et size, gi)-.11 E
-.165(ve)-.275 G 2.75(nt).165 G(hat the)-2.75 E(pack)100.8 383 Q
(et size is at most 1500 bytes.)-.11 E
(This does not mean that we need to change the TFRC-)5.5 E
(SP mechanisms for computing the allo)100.8 396 Q(wed transmit rate;)
-.275 E(this means simply that in)5.5 E -.275(eva)100.8 409 S
(luating the f).275 E(airness of TFRC-SP)-.11 E 2.75(,w)-1.221 G 2.75
(es)-2.75 G(hould consider f)-2.75 E(airness relati)-.11 E .33 -.165
(ve t)-.275 H 2.75(ot).165 G(he TCP \215o)-2.75 E(w)-.275 E
(using the optimal pack)100.8 422 Q
(et size \(though still at most 1500 bytes\) for that en)-.11 E
(vironment.)-.44 E/F2 11/Times-Bold@0 SF 2.75(4.5.2. Fragmentation)72
448 R(and the P)2.75 E(ath MTU)-.11 E F0(This document doesn')100.8
464.6 Q 2.75(ts)-.198 G(pecify TFRC-SP beha)-2.75 E
(vior in terms of pack)-.22 E(et fragmentation and)-.11 E -.165(Pa)100.8
477.6 S(th MTU Disco).165 E -.165(ve)-.165 G(ry \(PMTUD\).).165 E
(That is, should the transport protocol using TFRC-SP)5.5 E
(use PMTUD information to set an upper bound on the se)100.8 490.6 Q
(gment size?)-.165 E(Should the transport)5.5 E(protocol allo)100.8
503.6 Q 2.75(wp)-.275 G(ack)-2.75 E(ets to be fragmented in the netw)
-.11 E 2.75(ork? W)-.11 F 2.75(el)-.88 G(ea)-2.75 E .33 -.165(ve t)-.22
H(hese as questions for).165 E(the transport protocol.)100.8 516.6 Q
(As an e)5.5 E(xample, we note that DCCP requires that endpoints k)-.165
E(eep)-.11 E(track of the current PMTU, and says that fragmentation sho\
uld not be the def)100.8 529.6 Q(ault)-.11 E([RFC4340] \(Section 14\).)
100.8 542.6 Q F2 2.75(4.5.3. The)72 568.6 R
(Nominal Segment Size and the P)2.75 E(ath MTU)-.11 E F0
(When TFRC-SP is used with a nominal se)100.8 585.2 Q
(gment size s of 1460 bytes on a path where the)-.165 E(TCP MSS is in f)
100.8 598.2 Q(act only 536 bytes, the TFRC-SP \215o)-.11 E 2.75(wc)-.275
G(ould recei)-2.75 E .33 -.165(ve a)-.275 H(lmost three times the).165 E
(bandwidth, in bytes per second, as that of a TCP \215o)100.8 611.2 Q
2.75(wu)-.275 G(sing an MSS of 536 bytes.)-2.75 E(Similarly)100.8 624.2
Q 2.75(,i)-.715 G 2.75(na)-2.75 G 2.75(ne)-2.75 G -.44(nv)-2.75 G
(ironment with an MSS close to 4000 bytes, a TCP \215o).44 E 2.75(wc)
-.275 G(ould recei)-2.75 E -.165(ve)-.275 G
(almost three times the bandwidth of a TFRC-SP \215o)100.8 637.2 Q 2.75
(ww)-.275 G(ith its nominal se)-2.75 E(gment size s of)-.165 E
(1460 bytes.)100.8 650.2 Q(In both cases, we feel that these le)5.5 E
-.165(ve)-.275 G(ls of "unf).165 E(airness" with f)-.11 E(actors of tw)
-.11 E 2.75(oo)-.11 G(r)-2.75 E(three are acceptable; in particular)
100.8 663.2 Q 2.75(,t)-.44 G(he)-2.75 E 2.75(yw)-.165 G(on')-2.86 E 2.75
(tr)-.198 G(esult in one \215o)-2.75 E 2.75(wg)-.275 G
(rabbing all of the)-2.75 E -.22(av)100.8 676.2 S
(ailable bandwidth, to the e)-.055 E
(xclusion of the competing TCP or TFRC-SP \215o)-.165 E -.715(w.)-.275 G
(Flo)72 733 Q(yd/K)-.11 E 262.171(ohler Section)-.385 F 2.75(4.5.3. [P)
2.75 F(age 17])-.165 E 0 Cg EP
%%Page: 18 18
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E
(All IPv4 *end hosts* are required to accept and reassemble IP pack)
100.8 85 Q(ets of size 576 bytes)-.11 E([RFC791], b)100.8 98 Q
(ut IPv4 *links* do not necessarily ha)-.22 E .33 -.165(ve t)-.22 H 2.75
(os).165 G(upport this pack)-2.75 E(et size.)-.11 E(In slo)5.5 E(w)-.275
E(netw)100.8 111 Q(orks, the lar)-.11 E(gest possible pack)-.198 E
(et may tak)-.11 E 2.75(eac)-.11 G(onsiderable amount of time to send)
-2.75 E([RFC3819], and a smaller MTU may be desirable, e.g.)100.8 124 Q
(100')5.5 E 2.75(so)-.605 G 2.75(fb)-2.75 G 2.75(ytes. If)-2.75 F
(the \214rst-hop link)2.75 E(had a small MTU, then TCP w)100.8 137 Q
(ould choose an appropriately small MSS [RFC879].)-.11 E
([RFC1144] quotes cases of v)100.8 150 Q(ery lo)-.165 E 2.75(wl)-.275 G
(ink speeds were the MSS may be tens of bytes \(and)-2.75 E
(notes this is an e)100.8 163 Q(xtreme case\).)-.165 E 1.76 -.88(We n)
5.5 H(ote that if TFRC-SP is used o).88 E -.165(ve)-.165 G 2.75(rap).165
G(ath with an MTU)-2.75 E
(considerably smaller than 576 bytes, and the TFRC-SP \215o)100.8 176 Q
2.75(wu)-.275 G(ses a nominal se)-2.75 E(gment size s)-.165 E
(of 1460 bytes, then the TFRC-SP \215o)100.8 189 Q 2.75(wc)-.275 G
(ould recei)-2.75 E .33 -.165(ve c)-.275 H
(onsiderably more than three times).165 E
(the bandwidth of competing TCP \215o)100.8 202 Q(ws.)-.275 E
(If TFRC-SP is used with a nominal se)100.8 218.6 Q
(gment size s of less than 536 bytes \(becauses the path)-.165 E
(MTU is kno)100.8 231.6 Q
(wn to be less than 576 bytes\), then TFRC-SP is lik)-.275 E
(ely to be of minimal bene\214t)-.11 E(to applications.)100.8 244.6 Q
(If TFRC-SP w)5.5 E(as to be used on paths that ha)-.11 E .33 -.165
(ve a p)-.22 H(ath MTU of considerably).165 E
(less than 576 bytes, and the transport protocol w)100.8 257.6 Q
(as not required to k)-.11 E(eep track of the path)-.11 E
(MTU, this could result in the TFRC-SP \215o)100.8 270.6 Q 2.75(wu)-.275
G(sing the def)-2.75 E(ault nominal se)-.11 E(gment size of)-.165 E
(1460 bytes, and as a result recei)100.8 283.6 Q
(ving considerably more bandwidth than competing TCP)-.275 E<8d6f>100.8
296.6 Q 2.75(ws. As)-.275 F 2.75(ar)2.75 G(esult, TFRC-SP is not recomm\
ended for use with transport protocols that)-2.75 E(don')100.8 309.6 Q
2.75(tm)-.198 G(aintain some kno)-2.75 E(wledge of the path MTU.)-.275 E
/F1 11/Times-Bold@0 SF(4.6.)72 335.6 Q/F2 13/Times-Bold@0 SF
(The Loss Inter)5.5 E -.13(va)-.13 G 3.25(lL).13 G(ength f)-3.25 E
(or Short Loss Inter)-.325 E -.13(va)-.13 G(ls).13 E F0 -.165(Fo)100.8
352.2 S 2.75(raT).165 G(FRC-SP recei)-2.75 E -.165(ve)-.275 G .88 -.44
(r, t).165 H(he guidelines from Section 6 of RFC 3448 go).44 E -.165(ve)
-.165 G(rn when the).165 E(recei)100.8 365.2 Q -.165(ve)-.275 G 2.75(rs)
.165 G(hould send feedback messages.)-2.75 E(In particular)5.5 E 2.75
(,f)-.44 G(rom [RFC3448], "a feedback)-2.75 E(pack)100.8 378.2 Q
(et should ... be sent whene)-.11 E -.165(ve)-.275 G 2.75(ran).165 G .55
-.275(ew l)-2.75 H(oss e).275 E -.165(ve)-.275 G
(nt is detected without w).165 E(aiting for the end)-.11 E(of an R)100.8
391.2 Q 2.75(TT". In)-.66 F(addition, feedback pack)2.75 E
(ets are sent at least once per R)-.11 E(TT)-.66 E(.)-.814 E -.165(Fo)
100.8 407.8 S 2.75(raT).165 G
(FRC-SP connection with a short current loss interv)-2.75 E
(al \(less than tw)-.275 E 2.75(or)-.11 G(ound-trip)-2.75 E
(times\), it is possible for the loss interv)100.8 420.8 Q
(al length calculated for that loss interv)-.275 E(al to change in)-.275
E(odd w)100.8 433.8 Q(ays as additional pack)-.11 E
(et losses in that loss interv)-.11 E(al are detected.)-.275 E 1.76 -.88
(To p)5.5 H(re).88 E -.165(ve)-.275 G(nt).165 E
(unnecessary oscillations in the a)100.8 446.8 Q -.165(ve)-.22 G
(rage loss interv).165 E(al, Section 3 speci\214es that the current)
-.275 E(loss interv)100.8 459.8 Q
(al can be included in the calculation of the a)-.275 E -.165(ve)-.22 G
(rage loss interv).165 E(al only if the)-.275 E(current loss interv)
100.8 472.8 Q(al is longer than tw)-.275 E 2.75(or)-.11 G
(ound-trip times.)-2.75 E F1(5.)72 498.8 Q/F3 14/Times-Bold@0 SF 3.5(AC)
5.5 G(omparison with RFC 3714)-3.5 E F0
(RFC 3714 considers the problems of f)100.8 515.4 Q
(airness, potential congestion collapse, and poor user)-.11 E
(quality that could occur with the deplo)100.8 528.4 Q
(yment of non-congestion-controlled IP telephon)-.11 E(y)-.165 E -.165
(ove)100.8 541.4 S 2.75(rc).165 G(ongested best-ef)-2.75 E(fort netw)
-.275 E 2.75(orks. The)-.11 F(March 2004 document cites ongoing ef)2.75
E(forts in)-.275 E(the IETF)100.8 554.4 Q 2.75(,i)-.88 G(ncluding w)
-2.75 E(ork on TFRC and DCCP)-.11 E 5.5(.R)-1.221 G
(FC 3714 recommends that for best-ef)-5.5 E(fort)-.275 E(traf)100.8
567.4 Q(\214c with applications that ha)-.275 E .33 -.165(ve a \214)-.22
H -.165(xe).165 G 2.75(do).165 G 2.75(rm)-2.75 G
(inimum sending rate, the application or)-2.75 E
(transport protocol should monitor the pack)100.8 580.4 Q
(et drop rate, and discontinue sending for a)-.11 E
(period if the steady-state pack)100.8 593.4 Q
(et drop rate signi\214cantly e)-.11 E(xceeds the allo)-.165 E
(wed threshold for)-.275 E(that minimum sending rate.)100.8 606.4 Q
(In determining the allo)100.8 623 Q(wed pack)-.275 E
(et drop rate for a \214x)-.11 E(ed sending rate, RFC 3714 assumes)-.165
E(that an IP telephon)100.8 636 Q 2.75<798d>-.165 G .55 -.275(ow s)-2.75
H(hould be allo).275 E(wed to use the same sending rate in bytes per)
-.275 E(second as a 1500-byte-pack)100.8 649 Q(et TCP \215o)-.11 E 2.75
(we)-.275 G(xperiencing the same pack)-2.915 E(et drop rate as that of)
-.11 E(the IP telephon)100.8 662 Q 2.75<798d>-.165 G -.275(ow)-2.75 G
5.5(.A)-.44 G 2.75(sa)-5.5 G 2.75(ne)-2.75 G(xample, follo)-2.915 E
(wing this guideline, a V)-.275 E(oIP connection with a)-1.419 E
(round-trip time of 0.1 sec and a minimum sending rate of 64 Kbps w)
100.8 675 Q(ould be required to)-.11 E
(terminate or suspend when the persistent pack)100.8 688 Q
(et drop rate signi\214cantly e)-.11 E(xceeded 25%.)-.165 E(Flo)72 733 Q
(yd/K)-.11 E 278.671(ohler Section)-.385 F 2.75(5. [P)2.75 F(age 18])
-.165 E 0 Cg EP
%%Page: 19 19
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E(One lim\
itation of the lack of \214ne-grained control in the minimal mechanism \
described in)100.8 85 Q(RFC 3714 is that an IP telephon)100.8 98 Q 2.75
<798d>-.165 G .55 -.275(ow w)-2.75 H
(ould not adapt its sending rate in response to).165 E 2.75
(congestion. In)100.8 111 R(contrast, with TFRC-SP)2.75 E 2.75(,as)
-1.221 G(mall-pack)-2.75 E(et \215o)-.11 E 2.75(ww)-.275 G
(ould reduce its sending rate)-2.86 E(some)100.8 124 Q
(what in response to moderate pack)-.275 E(et drop rates, possibly a)
-.11 E -.22(vo)-.22 G(iding a period with).22 E(unnecessarily-hea)100.8
137 Q(vy pack)-.22 E(et drop rates in the netw)-.11 E(ork.)-.11 E
(Because RFC 3714 assumes that the allo)100.8 153.6 Q(wed pack)-.275 E
(et drop rate for an IP telephon)-.11 E 2.75<798d>-.165 G .55 -.275
(ow i)-2.75 H(s).275 E(determined by the sending rate that a TCP \215o)
100.8 166.6 Q 2.75(ww)-.275 G(ould use *with the same pack)-2.86 E
(et drop)-.11 E(rate*, the minimal mechanism in RFC 3714 w)100.8 179.6 Q
(ould not pro)-.11 E(vide f)-.165 E(airness between TCP and)-.11 E
(IP telephon)100.8 192.6 Q 2.75(yt)-.165 G(raf)-2.75 E(\214c in an en)
-.275 E(vironment where small pack)-.44 E(ets are less lik)-.11 E
(ely to be dropped)-.11 E(than lar)100.8 205.6 Q(ge pack)-.198 E 2.75
(ets. In)-.11 F(such an en)2.75 E(vironment, the small-pack)-.44 E
(et IP telephon)-.11 E 2.75<798d>-.165 G .55 -.275(ow w)-2.75 H(ould)
.165 E(mak)100.8 218.6 Q 2.75(et)-.11 G
(he optimistic assumption that a lar)-2.75 E(ge-pack)-.198 E
(et TCP \215o)-.11 E 2.75(ww)-.275 G(ould recei)-2.86 E .33 -.165(ve t)
-.275 H(he same).165 E(pack)100.8 231.6 Q
(et drop rate as the IP telephon)-.11 E 2.75<798d>-.165 G -.275(ow)-2.75
G 2.75(,a)-.44 G(nd as a result the small-pack)-2.75 E(et IP telephon)
-.11 E 2.75<798d>-.165 G -.275(ow)-2.75 G -.11(wo)100.8 244.6 S
(uld recei).11 E .33 -.165(ve a l)-.275 H(ar).165 E
(ger fraction of the link bandwidth than a competing lar)-.198 E
(ge-pack)-.198 E(et TCP)-.11 E<8d6f>100.8 257.6 Q -.715(w.)-.275 G/F1 11
/Times-Bold@0 SF(6.)72 283.6 Q/F2 14/Times-Bold@0 SF(TFRC-SP with A)5.5
E(pplications that Modify the P)-.35 E(ack)-.14 E(et Size)-.14 E F0
(One possible use for TFRC-SP w)100.8 300.2 Q
(ould be with applications that maintain a \214x)-.11 E(ed sending)-.165
E(rate in pack)100.8 313.2 Q(ets per second, b)-.11 E
(ut modify their pack)-.22 E(et size in response to congestion.)-.11 E
(TFRC-)5.5 E(SP monitors the connection')100.8 326.2 Q 2.75(sp)-.605 G
(ack)-2.75 E(et drop rate, and determines the allo)-.11 E
(wed sending rate in)-.275 E(bytes per second.)100.8 339.2 Q(Gi)5.5 E
-.165(ve)-.275 G 2.75(na).165 G 2.75(na)-2.75 G(pplication with a \214x)
-2.75 E(ed sending rate in pack)-.165 E(ets per second, the)-.11 E
(TFRC-SP sender could determine the data pack)100.8 352.2 Q
(et size that w)-.11 E(ould be needed for the sending)-.11 E
(rate in bytes per second not to e)100.8 365.2 Q(xceed the allo)-.165 E
(wed sending rate.)-.275 E(In en)5.5 E(vironments where the)-.44 E(pack)
100.8 378.2 Q(et drop rate is af)-.11 E(fected by the pack)-.275 E
(et size, decreasing the pack)-.11 E(et size could also result)-.11 E
(in a decrease in the pack)100.8 391.2 Q(et drop rate e)-.11 E
(xperienced by the \215o)-.165 E -.715(w.)-.275 G(There are man)100.8
407.8 Q 2.75(yq)-.165 G(uestions about ho)-2.75 E 2.75(wa)-.275 G 2.75
(na)-2.75 G(dapti)-2.75 E .33 -.165(ve a)-.275 H(pplication w).165 E
(ould use TFRC-SP that are)-.11 E(be)100.8 420.8 Q
(yond the scope of this document.)-.165 E(In particular)5.5 E 2.75(,a)
-.44 G 2.75(na)-2.75 G(pplication might wish to a)-2.75 E -.22(vo)-.22 G
(id).22 E(unnecessary reductions in the pack)100.8 433.8 Q(et size.)-.11
E(In this case, an application might w)8.25 E(ait for some)-.11 E
(period of time before reducing the pack)100.8 446.8 Q(et size, to a)
-.11 E -.22(vo)-.22 G(id an unnecessary reduction in the).22 E(pack)
100.8 459.8 Q(et size.)-.11 E(The details of ho)5.5 E 2.75(wl)-.275 G
(ong an application might w)-2.75 E(ait before reducing the pack)-.11 E
(et)-.11 E(size can be addressed in documents that are more application\
-speci\214c.)100.8 472.8 Q(Similarly)100.8 489.4 Q 2.75(,a)-.715 G 2.75
(na)-2.75 G(pplication using TFRC-SP might only ha)-2.75 E .33 -.165
(ve a f)-.22 H .55 -.275(ew p).165 H(ack).275 E
(et sizes that it is able)-.11 E(to use.)100.8 502.4 Q
(In this case, the application might not reduce the pack)5.5 E
(et size until the current)-.11 E(pack)100.8 515.4 Q
(et drop rate has signi\214cantly e)-.11 E(xceeded the pack)-.165 E
(et drop rate threshold for the current)-.11 E(sending rate, to a)100.8
528.4 Q -.22(vo)-.22 G(id unnecessary oscillations between tw).22 E 2.75
(op)-.11 G(ack)-2.75 E(et sizes and tw)-.11 E 2.75(os)-.11 G(ending)
-2.75 E 2.75(rates. Ag)100.8 541.4 R(ain, the details will ha)-.055 E
.33 -.165(ve t)-.22 H 2.75(ob).165 G 2.75(ea)-2.75 G
(ddressed in documents that are more application-)-2.75 E(speci\214c.)
100.8 554.4 Q(Because the allo)100.8 571 Q
(wed sending rate in TFRC-SP is in bytes per second rather than in pack)
-.275 E(ets)-.11 E(per second, there is little opportunity for applicat\
ions to manipulate the pack)100.8 584 Q(et size in order)-.11 E(to "g)
100.8 597 Q(ame" the system.)-.055 E(This dif)5.5 E
(fers from TFRC in CCID 3 \([RFC4342] \(Section 5.3\)\),)-.275 E
(where the allo)100.8 610 Q(wed sending rate is in pack)-.275 E
(ets per second.)-.11 E(In particular)5.5 E 2.75(,aT)-.44 G(FRC-SP)-2.75
E(application that sends small pack)100.8 623 Q(ets for a while, b)-.11
E(uilding up a f)-.22 E(ast sending rate, and then)-.11 E
(switches to lar)100.8 636 Q(ge pack)-.198 E(ets, w)-.11 E
(ould not increase its o)-.11 E -.165(ve)-.165 G
(rall sending rate by the change of).165 E(pack)100.8 649 Q(et size.)
-.11 E(Flo)72 733 Q(yd/K)-.11 E 278.671(ohler Section)-.385 F 2.75
(6. [P)2.75 F(age 19])-.165 E 0 Cg EP
%%Page: 20 20
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E/F1 11
/Times-Bold@0 SF(7.)72 85 Q/F2 14/Times-Bold@0 SF(Simulations)5.5 E F0(\
This section describes the performance of TFRC-SP in simulation scenari\
os with)100.8 101.6 Q(con\214gured pack)100.8 114.6 Q(et or byte drop r\
ates, and in scenarios with a range of queue management)-.11 E
(mechanisms at the congested link.)100.8 127.6 Q
(The simulations, described in detail in Appendix B,)5.5 E -.165(ex)
100.8 140.6 S(plore en).165 E(vironments where standard TFRC signi\214c\
antly limits the throughput of small-)-.44 E(pack)100.8 153.6 Q
(et \215o)-.11 E(ws, and TFRC-SP gi)-.275 E -.165(ve)-.275 G 2.75(st)
.165 G(he desired throughput.)-2.75 E(The simulations also e)5.5 E
(xplore)-.165 E(en)100.8 166.6 Q(vironments where standard TFRC allo)
-.44 E(ws small-pack)-.275 E(et \215o)-.11 E(ws to recei)-.275 E .33
-.165(ve g)-.275 H(ood).165 E(performance, while TFRC-SP is o)100.8
179.6 Q -.165(ve)-.165 G(rly aggressi).165 E -.165(ve)-.275 G(.).165 E
(The general lessons from the simulations are as follo)100.8 196.2 Q
(ws.)-.275 E 12.65<8349>100.8 212.8 S 2.75(ns)-12.65 G
(cenarios where lar)-2.75 E(ge and small pack)-.198 E(ets recei)-.11 E
.33 -.165(ve s)-.275 H(imilar pack).165 E(et drop rates, TFRC-SP)-.11 E
(gi)117.3 225.8 Q -.165(ve)-.275 G 2.75(sr).165 G
(oughly the desired sending rate \(Appendix B.1, B.2\).)-2.75 E 12.65
<8349>100.8 242.4 S 2.75(ns)-12.65 G
(cenarios where each *byte* is equally lik)-2.75 E
(ely to be dropped, standard TFRC gi)-.11 E -.165(ve)-.275 G(s).165 E
(reasonable f)117.3 255.4 Q(airness between small-pack)-.11 E
(et TFRC \215o)-.11 E(ws and lar)-.275 E(ge-pack)-.198 E(et TCP \215o)
-.11 E(ws)-.275 E(\(Appendix B.2\).)117.3 268.4 Q 12.65<8349>100.8 285 S
2.75(ns)-12.65 G(cenarios where small pack)-2.75 E(ets are less lik)-.11
E(ely to be dropped than lar)-.11 E(ge pack)-.198 E(ets,)-.11 E
(TFRC-SP does not gi)117.3 298 Q .33 -.165(ve r)-.275 H(easonable f).165
E(airness between small-pack)-.11 E(et TFRC-SP \215o)-.11 E(ws and)-.275
E(lar)117.3 311 Q(ge-pack)-.198 E(et TCP \215o)-.11 E(ws; small-pack)
-.275 E(et TFRC-SP \215o)-.11 E(ws can recei)-.275 E .33 -.165(ve c)
-.275 H(onsiderably more).165 E(bandwidth than competing lar)117.3 324 Q
(ge-pack)-.198 E(et TCP \215o)-.11 E(ws, and in some cases lar)-.275 E
(ge-pack)-.198 E(et)-.11 E(TCP \215o)117.3 337 Q
(ws can be essentially starv)-.275 E(ed by competing small-pack)-.165 E
(et TFRC-SP \215o)-.11 E(ws.)-.275 E(\(Appendix B.2, B.3, B.4\).)117.3
350 Q 12.65<8353>100.8 366.6 S(cenarios where small pack)-12.65 E
(ets are less lik)-.11 E(ely to be dropped than lar)-.11 E(ge pack)-.198
E(ets include)-.11 E(those with Drop-T)117.3 379.6 Q
(ail queues in bytes, and with A)-.88 E(QM mechanisms in byte mode)-.605
E(\(Appendix B.3, B.4\).)117.3 392.6 Q
(It has also been reported that wireless links are sometimes good)5.5 E
(enough to let small pack)117.3 405.6 Q(ets through, while lar)-.11 E
(ger pack)-.198 E(ets ha)-.11 E .33 -.165(ve a m)-.22 H
(uch higher error rate,).165 E
(and hence a higher drop probability [S05].)117.3 418.6 Q F1(8.)72 444.6
Q F2(General Discussion)5.5 E F0(Dropping rates for small pack)100.8
461.2 Q(ets: The goal of TFRC-SP is for small-pack)-.11 E
(et TFRC-SP \215o)-.11 E(ws)-.275 E(to ha)100.8 474.2 Q .33 -.165(ve r)
-.22 H(ough f).165 E(airness with lar)-.11 E(ge-pack)-.198 E
(et TCP \215o)-.11 E(ws in the sending rate in bps, in a scenario)-.275
E(where each pack)100.8 487.2 Q(et recei)-.11 E -.165(ve)-.275 G 2.75
(sr).165 G(oughly the same probability of being dropped.)-2.75 E
(In a scenario)5.5 E(where lar)100.8 500.2 Q(ge pack)-.198 E
(ets are more lik)-.11 E(ely to be dropped than small pack)-.11 E
(ets, or where \215o)-.11 E(ws with)-.275 E 2.75(ab)100.8 513.2 S
(ursty sending rate are more lik)-2.97 E(ely to ha)-.11 E .33 -.165
(ve p)-.22 H(ack).165 E(ets dropped than are \215o)-.11 E
(ws with a smooth)-.275 E(sending rate, small-pack)100.8 526.2 Q
(et TFRC-SP \215o)-.11 E(ws can recei)-.275 E .33 -.165(ve s)-.275 H
(igni\214cantly more bandwidth than).165 E(competing lar)100.8 539.2 Q
(ge-pack)-.198 E(et TCP \215o)-.11 E(ws.)-.275 E(The accurac)100.8 555.8
Q 2.75(yo)-.165 G 2.75(ft)-2.75 G
(he TCP response function used in TFRC: F)-2.75 E
(or applications with a)-.165 E
(maximum sending rate of 96 Kbps or less \(i.e., pack)100.8 568.8 Q
(ets of at most 120 bytes\) TFRC-SP)-.11 E
(only restricts the sending rate when the pack)100.8 581.8 Q
(et drop rate is f)-.11 E(airly high, e.g., greater than)-.11 E 2.75
(10%. [Deri)100.8 594.8 R -.275(va)-.275 G(tion: A TFRC-SP \215o).275 E
2.75(ww)-.275 G(ith a 200 ms round-trip time and a maximum)-2.75 E
(sending rate with pack)100.8 607.8 Q(et headers of 128 Kbps w)-.11 E
(ould ha)-.11 E .33 -.165(ve a s)-.22 H(ending rate in bytes per).165 E
(second equi)100.8 620.8 Q -.275(va)-.275 G(lent to a TCP \215o).275 E
2.75(ww)-.275 G(ith 1460-byte se)-2.75 E(gments sending 2.2 pack)-.165 E
(ets per round-)-.11 E(trip time.)100.8 633.8 Q(From T)5.5 E
(able 1 of RFC 3714, this sending rate can be sustained with a pack)-.88
E(et drop)-.11 E(rate slightly greater than 10%.])100.8 646.8 Q
(In this high-pack)5.5 E(et-drop re)-.11 E(gime, the performance of)
-.165 E(TFRC-SP is determined in part by the accurac)100.8 659.8 Q 2.75
(yo)-.165 G 2.75(ft)-2.75 G(he TCP response function in)-2.75 E
(representing the actual sending rate of a TCP connection.)100.8 672.8 Q
(Flo)72 733 Q(yd/K)-.11 E 278.671(ohler Section)-.385 F 2.75(8. [P)2.75
F(age 20])-.165 E 0 Cg EP
%%Page: 21 21
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E
(In the re)100.8 85 Q(gime of high pack)-.165 E
(et drop rates, TCP performance is also af)-.11 E(fected by the TCP)
-.275 E(algorithm \(e.g., SA)100.8 98 Q(CK or not\), by the minimum R)
-.44 E -.198(TO)-.66 G 2.75(,b).198 G 2.75(yt)-2.75 G
(he use or not of Limited)-2.75 E -.385(Tr)100.8 111 S
(ansmit, by the use of ECN, and the lik).385 E 2.75(e. It)-.11 F
(is good to ensure that simulations or)2.75 E -.165(ex)100.8 124 S
(periments e).165 E(xploring f)-.165 E(airness include the e)-.11 E
(xploration of f)-.165 E(airness with the most aggressi)-.11 E -.165(ve)
-.275 G(TCP mechanisms conformant with the current standards.)100.8 137
Q(Our simulations use SA)5.5 E(CK TCP)-.44 E(with Limited T)100.8 150 Q
(ransmit and with a minimum R)-.385 E .396 -.198(TO o)-.66 H 2.75(f2)
.198 G(00 ms.)-2.75 E(The simulation results are)5.5 E(lar)100.8 163 Q(\
gely the same with or without timestamps; timestamps were not used for \
simulations)-.198 E(reported in this paper)100.8 176 Q 5.5(.W)-.605 G
2.75(ed)-6.38 G(idn')-2.75 E 2.75(tu)-.198 G
(se TCP with ECN in setting the tar)-2.75 E(get sending rate for)-.198 E
(TFRC, because, as e)100.8 189 Q(xplained in Appendix B.1, our e)-.165 E
(xpectation is that in high pack)-.165 E(et drop)-.11 E(re)100.8 202 Q
(gimes, routers will drop rather than mark pack)-.165 E
(ets, either from polic)-.11 E 2.75(yo)-.165 G 2.75(rf)-2.75 G(rom b)
-2.75 E(uf)-.22 E(fer)-.275 E -.165(ove)100.8 215 S(r\215o).165 E -.715
(w.)-.275 G -.165(Fa)100.8 231.6 S(irness with dif).165 E(ferent pack)
-.275 E(et header sizes: In an en)-.11 E(vironment with IPv6 and/or se)
-.44 E -.165(ve)-.275 G(ral).165 E(layers of netw)100.8 244.6 Q
(ork-layer tunnels \(e.g., IPsec, GRE\), the pack)-.11 E
(et header could be 60, 80, or)-.11 E(100 bytes instead of the def)100.8
257.6 Q(ault 40 bytes assumed in Section 3.)-.11 E -.165(Fo)5.5 G 2.75
(ra).165 G 2.75(na)-2.75 G(pplication with)-2.75 E
(small ten-byte data se)100.8 270.6 Q
(gments, this means that the actual pack)-.165 E
(et size could be 70, 90, or 110)-.11 E(bytes, instead of the 50 bytes \
assumed by TFRC-SP in calculating the allo)100.8 283.6 Q
(wed sending rate.)-.275 E(Thus, a TFRC-SP application with lar)100.8
296.6 Q(ge headers could recei)-.198 E .33 -.165(ve m)-.275 H
(ore than twice the).165 E(bandwidth \(including the bandwidth used by \
headers\) than a TFRC-SP application with)100.8 309.6 Q(small headers.)
100.8 322.6 Q 1.76 -.88(We d)5.5 H 2.75(on).88 G(ot e)-2.75 E
(xpect this to be a problem; in particular)-.165 E 2.75(,T)-.44 G
(FRC-SP applications)-2.75 E(with lar)100.8 335.6 Q
(ge headers will not signi\214cantly de)-.198 E
(grade the performance of competing TCP)-.165 E
(applications, or of competing TFRC-SP applications with small headers.)
100.8 348.6 Q(General issues for TFRC: The congestion control mechanism\
s in TFRC and TFRC-SP)100.8 365.2 Q(limit a \215o)100.8 378.2 Q(w')-.275
E 2.75(ss)-.605 G(ending rate in pack)-2.75 E(ets per second.)-.11 E
(Simulations by T)5.5 E(om Phelan [P04])-.88 E -.165(ex)100.8 391.2 S
(plore ho).165 E 2.75(ws)-.275 G
(uch a limitation in sending rate can lead to unf)-2.75 E
(airness for the TFRC \215o)-.11 E 2.75(wi)-.275 G(n)-2.75 E
(some scenarios, e.g., when competing with b)100.8 404.2 Q
(ursty TCP web traf)-.22 E(\214c, in scenarios with lo)-.275 E(w)-.275 E
(le)100.8 417.2 Q -.165(ve)-.275 G(ls of statistical multiple).165 E
(xing at the congested link.)-.165 E/F1 11/Times-Bold@0 SF(9.)72 443.2 Q
/F2 14/Times-Bold@0 SF(Security Considerations)5.5 E F0
(There are no security considerations introduced in this document.)100.8
459.8 Q
(General security considerations for TFRC are discussed in RFC 3448.)
100.8 476.4 Q(The security)5.5 E
(considerations for TFRC include the need to protect ag)100.8 489.4 Q
(ainst spoofed feedback, and the)-.055 E(need for protection mechanisms\
 to protect the congestion control mechanisms ag)100.8 502.4 Q(ainst)
-.055 E(incorrect information from the recei)100.8 515.4 Q -.165(ve)
-.275 G -.605(r.).165 G(Security considerations for DCCP')100.8 532 Q
2.75(sC)-.605 G(ongestion Control ID 3, TFRC Congestion Control,)-2.75 E
(are discussed in [RFC4342].)100.8 545 Q(That document e)5.5 E(xtensi)
-.165 E -.165(ve)-.275 G(ly discussed the mechanisms the).165 E
(sender can use to v)100.8 558 Q
(erify the information sent by the recei)-.165 E -.165(ve)-.275 G -.605
(r.).165 G F1(10.)72 584 Q F2(IAN)5.5 E 3.5(AC)-.28 G(onsiderations)-3.5
E F0(There are no IAN)100.8 600.6 Q 2.75(Ac)-.385 G
(onsiderations in this document.)-2.75 E F1(11.)72 626.6 Q F2
(Conclusions)5.5 E F0(This document has speci\214ed TFRC-SP)100.8 643.2
Q 2.75(,aS)-1.221 G(mall-P)-2.75 E(ack)-.165 E(et \(SP\) v)-.11 E
(ariant of TFRC, designed for)-.275 E(applications that send small pack)
100.8 656.2 Q(ets, with at most a hundred pack)-.11 E(ets per second, b)
-.11 E(ut that)-.22 E
(desire the same sending rate in bps as a TCP connection e)100.8 669.2 Q
(xperiencing the same pack)-.165 E(et drop)-.11 E(rate b)100.8 682.2 Q
(ut sending pack)-.22 E(ets of 1500 bytes.)-.11 E
(TFRC-SP competes reasonably f)5.5 E(airly with lar)-.11 E(ge-)-.198 E
(pack)100.8 695.2 Q(et TCP and TFRC \215o)-.11 E(ws in en)-.275 E
(vironments where lar)-.44 E(ge-pack)-.198 E(et \215o)-.11 E
(ws and small-pack)-.275 E(et)-.11 E(Flo)72 733 Q(yd/K)-.11 E 273.171
(ohler Section)-.385 F 2.75(11. [P)2.75 F(age 21])-.165 E 0 Cg EP
%%Page: 22 22
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E<8d6f>
100.8 85 Q(ws e)-.275 E(xperience similar pack)-.165 E(et drop rates, b)
-.11 E(ut recei)-.22 E -.165(ve)-.275 G 2.75(sm).165 G
(ore than its share of the)-2.75 E(bandwidth in bps in en)100.8 98 Q
(vironments where small pack)-.44 E(ets are less lik)-.11 E
(ely to be dropped or)-.11 E(mark)100.8 111 Q(ed than are lar)-.11 E
(ge pack)-.198 E 2.75(ets. As)-.11 F 2.75(ar)2.75 G(esult, TFRC-SP is e)
-2.75 E(xperimental, and is not intended)-.165 E(for widespread deplo)
100.8 124 Q(yment at this time in the global Internet.)-.11 E
(In order to allo)100.8 140.6 Q 2.75(we)-.275 G
(xperimentation with TFRC-SP in the Datagram Congestion Control)-2.915 E
(Protocol \(DCCP\), an e)100.8 153.6 Q
(xperimental Congestion Control IDenti\214er \(CCID\) will be used,)
-.165 E(based on TFRC-SP b)100.8 166.6 Q
(ut speci\214ed in a separate document.)-.22 E/F1 11/Times-Bold@0 SF
(12.)72 192.6 Q/F2 14/Times-Bold@0 SF(Thanks)5.5 E F0 1.76 -.88(We t)
100.8 209.2 T(hank T).88 E
(om Phelan for discussions of TFRC-SP and for his paper e)-.88 E
(xploring the f)-.165 E(airness)-.11 E(between TCP and TFRC-SP \215o)
100.8 222.2 Q 2.75(ws. W)-.275 F 2.75(et)-.88 G
(hank Lars Eggert, Gorry F)-2.75 E(airhurst, Mark Handle)-.165 E -.715
(y,)-.165 G(Ladan Gharai, Richard Nelson, Colin Perkins, Pete Sholander)
100.8 235.2 Q 2.75(,M)-.44 G(agnus W)-2.75 E(esterlund, and)-.88 E(Joer)
100.8 248.2 Q 2.75(gW)-.198 G(idmer for feedback on earlier v)-3.19 E
(ersions of this draft.)-.165 E 1.76 -.88(We a)5.5 H(lso thank the DCCP)
.88 E -.88(Wo)100.8 261.2 S(rking Group for feedback and discussions.)
.88 E F1(A.)72 287.2 Q F2 -.35(Ap)5.5 G(pendix: Related W).35 E
(ork on Small-P)-1.05 E(ack)-.14 E(et V)-.14 E(ariants of TFRC)-1.288 E
F0(Other proposals for v)100.8 303.8 Q
(ariants of TFRC for applications with v)-.275 E(ariable pack)-.275 E
(et sizes include)-.11 E([WBL04] and [V00]. [V00] proposed that TFRC sh\
ould be modi\214ed so that \215o)100.8 316.8 Q(ws are not)-.275 E
(penalized by sending smaller pack)100.8 329.8 Q 2.75(ets. In)-.11 F
(particular)2.75 E 2.75(,[)-.44 G(V00] proposes using the path MTU in)
-2.75 E(the TCP-friendly equation, instead of the actual pack)100.8
342.8 Q(et size used by TFRC, and counting)-.11 E(the pack)100.8 355.8 Q
(et drop rate by using an estimation algorithm that aggre)-.11 E -.055
(ga)-.165 G(tes both pack).055 E(et drops and)-.11 E(recei)100.8 368.8 Q
-.165(ve)-.275 G 2.75(dp).165 G(ack)-2.75 E(ets into MTU-sized units.)
-.11 E([WBL04] also ar)100.8 385.4 Q(gues that adapting TFRC for v)-.198
E(ariable pack)-.275 E(et sizes by just using the pack)-.11 E(et)-.11 E
(size of a reference TCP \215o)100.8 398.4 Q 2.75(wi)-.275 G 2.75(nt)
-2.75 G(he TFRC equation w)-2.75 E(ould not suf)-.11 E
(\214ce in the high-pack)-.275 E(et-)-.11 E(loss re)100.8 411.4 Q
(gime; such a modi\214ed TFRC w)-.165 E(ould ha)-.11 E .33 -.165(ve a s)
-.22 H(trong bias in f).165 E -.22(avo)-.11 G 2.75(ro).22 G 2.75(fs)
-2.75 G(maller pack)-2.75 E(ets,)-.11 E(because multiple lost pack)100.8
424.4 Q(ets in a single round-trip time w)-.11 E(ould be aggre)-.11 E
-.055(ga)-.165 G(ted into a single).055 E(pack)100.8 437.4 Q(et loss.)
-.11 E([WBL04] proposes modifying the loss measurement process to accou\
nt for the)5.5 E(bias in f)100.8 450.4 Q -.22(avo)-.11 G 2.75(ro).22 G
2.75(fs)-2.75 G(maller pack)-2.75 E(ets.)-.11 E(The TFRC-SP v)100.8 467
Q(ariant of TFRC proposed in our document dif)-.275 E
(fers from [WBL04] in)-.275 E(restricting its attention to \215o)100.8
480 Q(ws that send at most 100 pack)-.275 E(ets per second.)-.11 E
(The TFRC-SP)5.5 E -.275(va)100.8 493 S
(riant proposed in our document also dif).275 E(fers from the stra)-.275
E 2.75(wp)-.165 G(roposal discussed in)-2.75 E([WBL04] in that the allo)
100.8 506 Q(wed bandwidth includes the bandwidth used by pack)-.275 E
(et headers.)-.11 E([WBL04] discusses four methods for modifying the lo\
ss measurement process,)100.8 522.6 Q("unbiasing", "virtual pack)100.8
535.6 Q(ets", "random sampling", and "Loss Insensiti)-.11 E .33 -.165
(ve P)-.275 H(eriod \(LIP\)).165 E 2.75(scaling". [WBL04])100.8 548.6 R
(\214nds only the second and third methods suf)2.75 E(\214ciently rob)
-.275 E(ust when the)-.22 E(netw)100.8 561.6 Q(ork drops pack)-.11 E
(ets independently of pack)-.11 E(et size.)-.11 E(The)5.5 E 2.75<798c>
-.165 G(nd only the second method)-2.75 E(suf)100.8 574.6 Q
(\214ciently rob)-.275 E(ust when the netw)-.22 E(ork is more lik)-.11 E
(ely to drop lar)-.11 E(ge pack)-.198 E(ets than small)-.11 E(pack)100.8
587.6 Q 2.75(ets. Our)-.11 F(method for calculating the loss e)2.75 E
-.165(ve)-.275 G(nt rate is some).165 E(what similar to the random)-.275
E(sampling method proposed in [WBL04], e)100.8 600.6 Q
(xcept that randomization is not used; instead of)-.165 E
(randomization, the e)100.8 613.6 Q(xact pack)-.165 E
(et loss rate is computed for short loss interv)-.11 E(als, and the)
-.275 E(standard loss e)100.8 626.6 Q -.165(ve)-.275 G
(nt rate calculation is used for longer loss interv).165 E 2.75
(als. [WBL04])-.275 F(includes)2.75 E(simulations with a Bernoulli loss\
 model, a Bernoulli loss model with a drop rate v)100.8 639.6 Q(arying)
-.275 E -.165(ove)100.8 652.6 S 2.75(rt).165 G(ime, and a Gilbert loss \
model, as well as more realistic simulations with a range of)-2.75 E
(TCP and TFRC \215o)100.8 665.6 Q(ws.)-.275 E
([WBL04] produces both a byte-mode and a pack)100.8 682.2 Q(et-mode v)
-.11 E(ariant of the TFRC transport)-.275 E(protocol, for connections o)
100.8 695.2 Q -.165(ve)-.165 G 2.75(rp).165 G(aths with per)-2.75 E
(-byte and per)-.22 E(-pack)-.22 E(et dropping respecti)-.11 E -.165(ve)
-.275 G(ly).165 E(.)-.715 E(Flo)72 733 Q(yd/K)-.11 E 276.229
(ohler Section)-.385 F 2.75(A. [P)2.75 F(age 22])-.165 E 0 Cg EP
%%Page: 23 23
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E 1.76
-.88(We w)100.8 85 T(ould ar).77 E
(gue that in the absence of transport-le)-.198 E -.165(ve)-.275 G 2.75
(lm).165 G(echanisms for determining whether)-2.75 E(the pack)100.8 98 Q
(et dropping in the netw)-.11 E(ork is per)-.11 E(-pack)-.22 E(et, per)
-.11 E(-byte, or some)-.22 E(where in between, a)-.275 E
(single TFRC implementation is needed, independently of the pack)100.8
111 Q(et-dropping beha)-.11 E(viors of)-.22 E
(the routers along the path.)100.8 124 Q(It w)5.5 E
(ould of course be preferable to ha)-.11 E .33 -.165(ve a m)-.22 H
(echanism that gi).165 E -.165(ve)-.275 G(s).165 E
(roughly acceptable beha)100.8 137 Q(vior)-.22 E 2.75(,t)-.44 G 2.75(ot)
-2.75 G(he connection and to the netw)-2.75 E(ork as a whole, on paths)
-.11 E(with both per)100.8 150 Q(-byte and per)-.22 E(-pack)-.22 E
(et dropping \(and on paths with multiple congested routers,)-.11 E
(some with per)100.8 163 Q(-byte dropping mechanisms, some with per)-.22
E(-pack)-.22 E(et dropping mechanisms,)-.11 E
(and some with dropping mechanisms that lie some)100.8 176 Q
(where between per)-.275 E(-byte and per)-.22 E(-pack)-.22 E(et\).)-.11
E(An important contrib)100.8 192.6 Q(ution w)-.22 E(ould be to in)-.11 E
-.165(ve)-.44 G(stig).165 E(ate the range of beha)-.055 E
(viors actually present)-.22 E(in today')100.8 205.6 Q 2.75(sn)-.605 G
(etw)-2.75 E(orks, in terms of pack)-.11 E
(et-dropping as a function of pack)-.11 E(et size.)-.11 E/F1 11
/Times-Bold@0 SF(B.)72 231.6 Q/F2 14/Times-Bold@0 SF(Simulation Results)
5.5 E F0(This appendix reports on the simulation results outlined in Se\
ction 7 . TFRC-SP has been)100.8 248.2 Q(added to the NS simulator)100.8
261.2 Q 2.75(,a)-.44 G(nd is illustrated in the v)-2.75 E
(alidation test "./test-all-friendly" in the)-.275 E
(directory tcl/tests.)100.8 274.2 Q
(The simulation scripts and graphs for the simulations in this document)
5.5 E(are a)100.8 287.2 Q -.275(va)-.22 G(ilable at [V).275 E(OIPSIMS].)
-.44 E F1(B.1.)72 313.2 Q/F3 13/Times-Bold@0 SF
(Simulations with Con\214gur)5.5 E(ed P)-.234 E(ack)-.13 E(et Dr)-.13 E
(op Rates)-.234 E F0(In this section we describe simulation results fro\
m simulations comparing the throughput)100.8 329.8 Q(of standard \(SA)
100.8 342.8 Q(CK\) TCP \215o)-.44 E(ws, TCP \215o)-.275 E
(ws with timestamps and ECN, TFRC-SP \215o)-.275 E(ws,)-.275 E
(and standard TFRC \(Stnd TFRC\) \215o)100.8 355.8 Q 2.75(ws. In)-.275 F
(these simulations we con\214gure the router to)2.75 E
(randomly drop or mark pack)100.8 368.8 Q
(ets at a speci\214ed rate, independently of the pack)-.11 E(et size.)
-.11 E -.165(Fo)5.5 G(r).165 E(each speci\214ed pack)100.8 381.8 Q
(et drop rate, we gi)-.11 E .33 -.165(ve a \215)-.275 H -.275(ow).165 G
1.21 -.605('s a).275 H -.165(ve).385 G(rage sending rate in Kbps o).165
E -.165(ve)-.165 G 2.75(rt).165 G(he)-2.75 E
(second half of a 100-second simulation, a)100.8 394.8 Q -.165(ve)-.22 G
(raged o).165 E -.165(ve)-.165 G 2.75(rt).165 G(en \215o)-2.75 E(ws.)
-.275 E(Flo)72 733 Q(yd/K)-.11 E 268.584(ohler Section)-.385 F 2.75
(B.1. [P)2.75 F(age 23])-.165 E 0 Cg EP
%%Page: 24 24
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E/F1 11
/Courier@0 SF 39.6(Packet Send)180 85 R(Rates \(Kbps, 1460B seg\))6.6 E
33(DropRate TCP ECN)180 98 R 33(TCP TFRC)6.6 F 19.8(-------- --------)
180 111 R 13.2(-------- --------)19.8 F 19.8(0.001 2020.85 1904.61)199.8
124 R(982.09)33 E 26.4(0.005 811.10 792.11 878.08)199.8 137 R 33
(0.01 515.45)199.8 150 R 26.4(533.19 598.90)33 F 33(0.02 362.93)199.8
163 R 26.4(382.67 431.41)33 F 33(0.04 250.06)199.8 176 R 26.4
(252.64 284.82)33 F 33(0.05 204.48)199.8 189 R 26.4(218.16 268.51)33 F
39.6(0.1 143.30)199.8 202 R 26.4(148.41 146.03)33 F 46.2(0.2 78.65)199.8
215 R 26.4(93.23* 55.14)39.6 F 46.2(0.3 26.26)199.8 228 R 26.4
(59.65* 32.87)39.6 F 52.8(0.4 9.87)199.8 241 R 26.4(47.79* 25.45)39.6 F
52.8(0.5 3.53)199.8 254 R 26.4(32.01* 18.52)39.6 F 6.6(*E)133.8 280 S
(CN scenarios marked with an asterisk are not realistic,)-6.6 E
(as routers are not recommended to mark packets when packet)147 293 Q
(drop/mark rates are 20% or higher.)147 306 Q
(Table 7: Send Rate vs. Packet Drop Rate I:)180 332 Q
(1460B TFRC Segments)272.4 345 Q
(\(1.184 Kbps Maximum TFRC Data Sending Rate\))193.2 358 Q F0 -.88(Ta)
100.8 377 S(ble 7 sho).88 E
(ws the sending rate for a TCP and a standard TFRC \215o)-.275 E 2.75
(wf)-.275 G(or a range of)-2.75 E(con\214gured pack)100.8 390 Q
(et drop rates, when both \215o)-.11 E(ws ha)-.275 E .33 -.165(ve 1)-.22
H(460-byte data se).165 E(gments, in order to)-.165 E
(illustrate the relati)100.8 403 Q .33 -.165(ve f)-.275 H
(airness of TCP and TFRC when both \215o).055 E(ws use the same pack)
-.275 E(et size.)-.11 E -.165(Fo)100.8 416 S 2.75(re).165 G
(xample, a pack)-2.915 E
(et drop rate of 0.1 means that 10% of the TCP and TFRC pack)-.11 E
(ets are)-.11 E 2.75(dropped. The)100.8 429 R(TFRC \215o)2.75 E 2.75(wi)
-.275 G 2.75(sc)-2.75 G(on\214gured to send at most 100 pack)-2.75 E
(ets per second.)-.11 E(There is)5.5 E(good relati)100.8 442 Q .33 -.165
(ve f)-.275 H(airness until the pack).055 E
(et drop percentages reach 40 and 50%, when the TFRC)-.11 E<8d6f>100.8
455 Q 2.75(wr)-.275 G(ecei)-2.75 E -.165(ve)-.275 G 2.75(st).165 G
(hree to \214v)-2.75 E 2.75(et)-.165 G
(imes more bandwidth than the standard TCP \215o)-2.75 E 4.18 -.715
(w. W)-.275 H 2.75(en)-.165 G(ote that)-2.75 E(an ECN TCP \215o)100.8
468 Q 2.75(ww)-.275 G(ould recei)-2.86 E .33 -.165(ve a h)-.275 H
(igher throughput than the TFRC \215o).165 E 2.75(wa)-.275 G 2.75(tt)
-2.75 G(hese high)-2.75 E(pack)100.8 481 Q
(et drop rate, if ECN-marking w)-.11 E(as still being instead of pack)
-.11 E 2.75(et-dropping. Ho)-.11 F(we)-.275 E -.165(ve)-.275 G .88 -.44
(r, w).165 H(e).44 E(don')100.8 494 Q 2.75(tu)-.198 G
(se the ECN TCP sending rate in these high-pack)-2.75 E
(et-drop scenarios as the tar)-.11 E(get)-.198 E(sending rate for TFRC,\
 as routers are advised to drop rather than mark pack)100.8 507 Q
(ets during high)-.11 E(le)100.8 520 Q -.165(ve)-.275 G
(ls of congestion [RFC3168] \(Section 7\).).165 E
(In addition, there is lik)5.5 E(ely to be b)-.11 E(uf)-.22 E(fer)-.275
E -.165(ove)100.8 533 S(r\215o).165 E 2.75(wi)-.275 G 2.75(ns)-2.75 G
(cenarios with such high pack)-2.75 E
(et dropping/marking rates, forcing routers to drop)-.11 E(pack)100.8
546 Q(ets instead of ECN-marking them.)-.11 E(Flo)72 733 Q(yd/K)-.11 E
268.584(ohler Section)-.385 F 2.75(B.1. [P)2.75 F(age 24])-.165 E 0 Cg
EP
%%Page: 25 25
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E/F1 11
/Courier@0 SF 6.6(<------S)206.4 85 S(end Rates \(Kbps\) - - - - - >)
-6.6 E 39.6(Packet TCP ECN)147 98 R 26.4(TCP TFRC-SP)6.6 F(Stnd TFRC)
19.8 E 6.6(DropRate \(1460B)140.4 111 R(seg\) \(1460B seg\))6.6 E
(\(14B seg\))13.2 E(\(14B seg\))13.2 E 6.6(-------- -----------)140.4
124 R 6.6(----------- --------- ---------)6.6 F 19.8(0.001 1787.54)160.2
137 R 33(1993.03 17.71 17.69)33 F 26.4(0.005 785.11)160.2 150 R 33
(823.75 18.11 17.69)39.6 F 33(0.01 533.38 529.01 17.69 17.80)160.2 163 R
33(0.02 317.16 399.62 17.69 13.41)160.2 176 R 33
(0.04 245.42 260.57 17.69)160.2 189 R(8.84)46.2 E 33
(0.05 216.38 223.75 17.69)160.2 202 R(7.63)46.2 E 39.6(0.1 142.75)160.2
215 R 33(138.36 17.69)39.6 F(4.29)46.2 E 46.2(0.2 58.61)160.2 228 R 26.4
(91.54* 17.80)46.2 F(1.94)46.2 E 46.2(0.3 21.62)160.2 241 R 26.4
(63.96* 10.26)46.2 F(1.00)46.2 E 46.2(0.4 10.51)160.2 254 R 33
(41.74* 4.78)46.2 F(0.77)46.2 E 52.8(0.5 1.92)160.2 267 R 33
(19.03* 2.41)46.2 F(0.56)46.2 E 6.6(*E)133.8 293 S
(CN scenarios marked with an asterisk are not realistic,)-6.6 E
(as routers are not recommended to mark packets when packet)147 306 Q
(drop/mark rates are 20% or higher.)147 319 Q
(Table 8: Send Rate vs. Packet Drop Rate II:)180 345 Q
(14B TFRC Segments)279 358 Q
(\(5.6 Kbps Maximum TFRC Data Sending Rate\))199.8 371 Q F0 -.88(Ta)
100.8 393.6 S(ble 8 sho).88 E
(ws the results of simulations where each TFRC-SP \215o)-.275 E 2.75(wh)
-.275 G(as a maximum data)-2.75 E
(sending rate of 5.6 Kbps, with 14-byte data pack)100.8 406.6 Q
(ets and a 32-byte pack)-.11 E(et header for DCCP)-.11 E(and IP)100.8
419.6 Q 5.5(.E)-1.221 G(ach TCP \215o)-5.5 E 2.75(wh)-.275 G(as a recei)
-2.75 E .33 -.165(ve w)-.275 H(indo).165 E 2.75(wo)-.275 G 2.75(f1)-2.75
G(00 pack)-2.75 E(ets and a data pack)-.11 E(et size of 1460)-.11 E
(bytes, with a 40-byte pack)100.8 432.6 Q(et header for TCP and IP)-.11
E 5.5(.T)-1.221 G(he TCP \215o)-5.5 E 2.75(wu)-.275 G(ses SA)-2.75 E
(CK TCP with)-.44 E(Limited T)100.8 445.6 Q(ransmit, b)-.385 E
(ut without timestamps or ECN.)-.22 E(Each \215o)5.5 E 2.75(wh)-.275 G
(as a round-trip time of 240)-2.75 E
(ms in the absence of queueing delay)100.8 458.6 Q(.)-.715 E
(The TFRC sending rate in T)100.8 475.2 Q
(able 8 is the sending rate for the 14-byte data pack)-.88 E
(et with the)-.11 E(32-byte pack)100.8 488.2 Q(et header)-.11 E 5.5(.T)
-.605 G(hus, only 30% of the TFRC sending rate is for data, and with a)
-5.5 E(pack)100.8 501.2 Q
(et drop rate of p, only a fraction 1-p of that data mak)-.11 E
(es it to the recei)-.11 E -.165(ve)-.275 G 3.96 -.605(r. T).165 H
(hus, the).605 E(TFRC data recei)100.8 514.2 Q .33 -.165(ve r)-.275 H
(ate can be considerably less than the TFRC sending rate in the table.)
.165 E(Because TCP uses lar)100.8 527.2 Q(ge pack)-.198 E
(ets, 97% of the TCP sending rate is for data, and the same)-.11 E
(fraction 1-p of that data mak)100.8 540.2 Q(es it to the recei)-.11 E
-.165(ve)-.275 G -.605(r.).165 G -.88(Ta)100.8 556.8 S(ble 8 sho).88 E(\
ws that for the 5.6 Kbps data stream with TFRC, Standard TFRC \(Stnd TF\
RC\))-.275 E(gi)100.8 569.8 Q -.165(ve)-.275 G 2.75(sav).165 G
(ery poor sending rate in bps, relati)-2.915 E .33 -.165(ve t)-.275 H
2.75(ot).165 G(he sending rate for the lar)-2.75 E(ge-pack)-.198 E
(et TCP)-.11 E 2.75(connection. In)100.8 582.8 R
(contrast, the sending rate for the TFRC-SP \215o)2.75 E 2.75(wi)-.275 G
2.75(sr)-2.75 G(elati)-2.75 E -.165(ve)-.275 G(ly close to the).165 E
(desired goal of f)100.8 595.8 Q(airness in bps with TCP)-.11 E(.)-1.221
E -.88(Ta)100.8 612.4 S(ble 8 sho).88 E(ws that with TFRC-SP)-.275 E
2.75(,t)-1.221 G(he 5.6 Kbps data stream doesn')-2.75 E 2.75(tr)-.198 G
(educe its sending rate)-2.75 E(until pack)100.8 625.4 Q
(et drop rates greater than 20%, as desired.)-.11 E -.44(Wi)5.5 G
(th pack).44 E(et drop rates of 30-40%, the)-.11 E
(sending rate for the TFRC-SP \215o)100.8 638.4 Q 2.75(wi)-.275 G 2.75
(ss)-2.75 G(ome)-2.75 E(what less than that of the a)-.275 E -.165(ve)
-.22 G(rage lar).165 E(ge-pack)-.198 E(et)-.11 E(TCP \215o)100.8 651.4 Q
1.43 -.715(w, w)-.275 H(hile for pack).715 E
(et drop rates of 50% the sending rate for the TFRC-SP \215o)-.11 E 2.75
(wi)-.275 G(s)-2.75 E(some)100.8 664.4 Q
(what greater than that of the a)-.275 E -.165(ve)-.22 G(rage lar).165 E
(ge-pack)-.198 E(et TCP \215o)-.11 E -.715(w.)-.275 G(Flo)72 733 Q(yd/K)
-.11 E 268.584(ohler Section)-.385 F 2.75(B.1. [P)2.75 F(age 25])-.165 E
0 Cg EP
%%Page: 26 26
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E/F1 11
/Courier@0 SF 6.6(<------S)206.4 85 S(end Rates \(Kbps\) - - - - - >)
-6.6 E 39.6(Packet TCP ECN)147 98 R 19.8(TCP TFRC-SP)6.6 F(Stnd TFRC)
19.8 E 6.6(DropRate \(1460B)140.4 111 R
(seg\) \(1460B seg\) \(200B seg\) \(200B seg\))6.6 E 6.6
(-------- -----------)140.4 124 R(----------- ---------- ----------)6.6
E 19.8(0.001 1908.98)160.2 137 R 26.4(1869.24 183.45 178.35)33 F 26.4
(0.005 854.69)160.2 150 R 26.4(835.10 185.06 138.06)39.6 F 33
(0.01 564.10 531.10)160.2 163 R 33(185.33 92.43)33 F 33
(0.02 365.38 369.10)160.2 176 R 33(185.57 62.18)33 F 33
(0.04 220.80 257.81)160.2 189 R 33(185.14 45.43)33 F 33
(0.05 208.97 219.41)160.2 202 R 33(180.08 39.44)33 F 39.6(0.1 141.67)
160.2 215 R 26.4(143.88 127.33)39.6 F(21.96)39.6 E 46.2(0.2 62.66)160.2
228 R 26.4(91.87* 54.66)46.2 F(9.40)46.2 E 46.2(0.3 16.63)160.2 241 R
26.4(65.52* 24.50)46.2 F(4.73)46.2 E 52.8(0.4 6.62)160.2 254 R 26.4
(42.00* 13.47)46.2 F(3.35)46.2 E 52.8(0.5 1.01)160.2 267 R 26.4
(21.34* 10.51)46.2 F(2.92)46.2 E 6.6(*E)133.8 293 S
(CN scenarios marked with an asterisk are not realistic,)-6.6 E
(as routers are not recommended to mark packets when packet)147 306 Q
(drop/mark rates are 20% or higher.)147 319 Q
(Table 9: Sending Rate vs. Packet Drop Rate III:)180 345 Q
(200B TFRC Segments)265.8 358 Q
(\(160 Kbps Maximum TFRC Data Sending Rate\))186.6 371 Q F0 -.88(Ta)
100.8 390 S(ble 9 sho).88 E(ws results with con\214gured pack)-.275 E
(et drop rates when the TFRC \215o)-.11 E 2.75(wu)-.275 G(ses 200-byte)
-2.75 E(data pack)100.8 403 Q
(ets, with a maximum data sending rate of 160 Kbps.)-.11 E(As in T)5.5 E
(able 8, the)-.88 E(performance of Standard TFRC is quite poor)100.8 416
Q 2.75(,w)-.44 G(hile the performance of TFRC-SP is)-2.75 E
(essentially as desired for pack)100.8 429 Q(et drop rates up to 30%.)
-.11 E(Ag)5.5 E(ain as e)-.055 E(xpected, with pack)-.165 E(et drop)-.11
E(rates of 40-50% the TFRC-SP sending rate is some)100.8 442 Q
(what higher than desired.)-.275 E -.165(Fo)100.8 458.6 S 2.75(rt).165 G
(hese simulations, the sending rate of a TCP connection using timestamp\
s is similar to)-2.75 E(the sending rate sho)100.8 471.6 Q
(wn for a standard TCP connection without timestamps.)-.275 E/F2 11
/Times-Bold@0 SF(B.2.)72 497.6 Q/F3 13/Times-Bold@0 SF
(Simulations with Con\214gur)5.5 E(ed Byte Dr)-.234 E(op Rates)-.234 E
F0(In this section we e)100.8 514.2 Q(xplore simulations where the rout\
er is con\214gured to drop or mark each)-.165 E
(*byte* at a speci\214ed rate.)100.8 527.2 Q
(When a byte is chosen to be dropped \(or mark)5.5 E(ed\), the entire)
-.11 E(pack)100.8 540.2 Q(et containing that byte is dropped \(or mark)
-.11 E(ed\).)-.11 E(Flo)72 733 Q(yd/K)-.11 E 268.584(ohler Section)-.385
F 2.75(B.2. [P)2.75 F(age 26])-.165 E 0 Cg EP
%%Page: 27 27
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E/F1 11
/Courier@0 SF 6.6(<-----S)259.2 85 S(end Rates \(Kbps\) - - - - - >)-6.6
E 39.6(Byte TCP)133.8 98 R 19.8(TFRC-SP Stnd)178.2 F(TFRC)6.6 E 13.2
(DropRate SegSize)120.6 111 R 33(TCP ECN)33 F 19.8(TCP \(14B)6.6 F 6.6
(seg\) \(14B)6.6 F(seg\))6.6 E 13.2
(-------- ------- -------- -------- ---------)120.6 124 R(---------)13.2
E 26.4(0.00001 1460 423.02 431.26)127.2 137 R 33(17.69 17.69)39.6 F 33
(0.0001 1460)127.2 150 R 26.4(117.41 114.34)33 F 33(17.69 17.69)39.6 F
46.2(0.001 128)127.2 163 R 33(10.78 11.50 17.69)39.6 F(8.37)46.2 E 52.8
(0.005 14)127.2 176 R 39.6(1.75 2.89)46.2 F 39.6(18.39 1.91)39.6 F 39.6
(0.010 1460 0.31 0.26 7.07 0.84)127.2 189 R 39.6
(0.020 1460 0.29 0.26 1.61 0.43)127.2 202 R 39.6(0.040 1460 0.12 0.26*)
127.2 215 R 39.6(0.17 0.12)39.6 F 39.6(0.050 1460 0.15 0.26*)127.2 228 R
39.6(0.08 0.06)39.6 F 6.6(*E)133.8 254 S
(CN scenarios marked with an asterisk are not realistic,)-6.6 E
(as routers are not recommended to mark packets when packet)147 267 Q
(drop/mark rates are 20% or higher.)147 280 Q
(TFRC's maximum data sending rate is 5.6 Kbps.)147 306 Q
(Table 10: Sending Rate vs. Byte Drop Rate)153.6 332 Q F0 -.88(Ta)100.8
354.6 S(ble 10 sho).88 E(ws the TCP and TFRC send rates for v)-.275 E
(arious byte drop rates.)-.275 E -.165(Fo)5.5 G 2.75(re).165 G(ach byte)
-2.75 E(drop rate, T)100.8 367.6 Q(able 10 sho)-.88 E
(ws the TCP sending rate, with and without ECN, for the TCP)-.275 E(se)
100.8 380.6 Q(gment size that gi)-.165 E -.165(ve)-.275 G 2.75(st).165 G
(he highest TCP sending rate.)-2.75 E(Simulations were run with TCP)5.5
E(se)100.8 393.6 Q(gments of 14, 128, 256, 512, and 1460 bytes.)-.165 E
(Thus, for a byte drop rate of 0.00001, the)5.5 E(table sho)100.8 406.6
Q(ws the TCP sending rate with 1460-byte data se)-.275 E(gments, b)-.165
E(ut with a byte drop rate)-.22 E(of 0.001, the table sho)100.8 419.6 Q
(ws the TCP sending rate with 128-byte data se)-.275 E 2.75(gments. F)
-.165 F(or each byte)-.165 E(drop rate, T)100.8 432.6 Q
(able 10 also sho)-.88 E
(ws the TFRC-SP and that Standard TFRC sending rates.)-.275 E -.44(Wi)
5.5 G(th).44 E(con\214gured byte drop rates, TFRC-SP gi)100.8 445.6 Q
-.165(ve)-.275 G 2.75(sa).165 G 2.75(nu)-2.75 G(nf)-2.75 E(air adv)-.11
E(antage to the TFRC-SP \215o)-.275 E -.715(w,)-.275 G
(while Standard TFRC gi)100.8 458.6 Q -.165(ve)-.275 G 2.75(se).165 G
(ssentially the desired performance.)-2.75 E F1(TCP Pkt)232.8 477.6 Q
(TFRC Pkt)33 E 26.4(Byte DropRate)173.4 490.6 R 39.6(DropRate TCP/TFRC)
26.4 F 6.6(DropRate \(1460B)160.2 503.6 R 6.6(seg\) \(14B)6.6 F 13.2
(seg\) Pkt)6.6 F(Drop Ratio)6.6 E 6.6(-------- ----------- ---------)
160.2 516.6 R(--------------)19.8 E 26.4(0.00001 0.015)166.8 529.6 R
46.2(0.0006 26.59)46.2 F 33(0.0001 0.13)166.8 542.6 R 46.2(0.0056 24.94)
52.8 F 39.6(0.001 0.77)166.8 555.6 R 52.8(0.054 14.26)52.8 F 39.6
(0.005 0.99)166.8 568.6 R 66(0.24 4.08)52.8 F 46.2(0.01 1.00 0.43)166.8
581.6 R(2.32)72.6 E 46.2(0.05 1.00 0.94)166.8 594.6 R(1.05)72.6 E
(Table 11: Packet Drop Rate Ratio vs. Byte Drop Rate)153.6 620.6 Q F0
-.88(Ta)100.8 643.2 S(ble 11 con).88 E -.165(ve)-.44 G
(rts the byte drop rate p to pack).165 E
(et drop rates for the TCP and TFRC pack)-.11 E(ets,)-.11 E
(where the pack)100.8 656.2 Q(et drop rate for an N-byte pack)-.11 E
(et is 1-\(1-p\)^N.)-.11 E(Thus, a byte drop rate of)5.5 E
(0.001, with each byte being dropped with probability 0.001, con)100.8
669.2 Q -.165(ve)-.44 G(rts to a pack).165 E(et drop rate)-.11 E
(of 0.77, or 77%, for the 1500-byte TCP pack)100.8 682.2 Q
(ets, and a pack)-.11 E(et drop rate of 0.054, or 5.4%,)-.11 E
(for the 56-byte TFRC pack)100.8 695.2 Q(ets.)-.11 E(Flo)72 733 Q(yd/K)
-.11 E 268.584(ohler Section)-.385 F 2.75(B.2. [P)2.75 F(age 27])-.165 E
0 Cg EP
%%Page: 28 28
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E
(The right column of T)100.8 85 Q(able 11 sho)-.88 E
(ws the ratio between the TCP pack)-.275 E(et drop rate and the)-.11 E
(TFRC pack)100.8 98 Q(et drop rate.)-.11 E -.165(Fo)5.5 G 2.75(rl).165 G
.55 -.275(ow b)-2.75 H
(yte drop rates, this ratio is close to 26.8, the ratio).275 E
(between the TCP and TFRC pack)100.8 111 Q(et sizes.)-.11 E -.165(Fo)5.5
G 2.75(rh).165 G(igh byte drop rates, where e)-2.75 E -.165(ve)-.275 G
2.75(nm).165 G(ost small)-2.75 E(TFRC pack)100.8 124 Q(ets are lik)-.11
E(ely to be dropped, this drop ratio approaches 1.)-.11 E(As T)5.5 E
(able 10 sho)-.88 E(ws,)-.275 E(with byte drop rates, the Standard TFRC\
 sending rate is close to optimal, competing f)100.8 137 Q(airly)-.11 E
(with a TCP connection using the optimal pack)100.8 150 Q
(et size within the allo)-.11 E(wed range.)-.275 E(In contrast,)5.5 E(t\
he TFRC-SP connection gets more than its share of the bandwidth, though\
 it does reduce)100.8 163 Q(its sending rate for a byte drop rate of 0.\
01 or more \(corresponding to a TFRC-SP *pack)100.8 176 Q(et*)-.11 E
(drop rate of 0.43.)100.8 189 Q -.88(Ta)100.8 205.6 S
(ble 10 essentially sho).88 E(ws three separate re)-.275 E 2.75
(gions. In)-.165 F(the lo)2.75 E(w-congestion re)-.275 E
(gion, with byte)-.165 E(drop rates less than 0.0001, the TFRC-SP conne\
ction is sending at its maximum sending)100.8 218.6 Q 2.75(rate. In)
100.8 231.6 R(this re)2.75 E
(gion the optimal TCP connection is the one with 1460-byte se)-.165 E
(gments, and)-.165 E(the TCP sending rate goes as 1/sqrt\(p\), for pack)
100.8 244.6 Q(et drop rate p.)-.11 E(This lo)5.5 E(w-congestion re)-.275
E(gion)-.165 E(holds for pack)100.8 257.6 Q(et drop rates up to 10-15%.)
-.11 E(In the middle re)100.8 274.2 Q(gion of T)-.165 E
(able 10, with byte drop rates from 0.0001 to 0.005, the optimal)-.88 E
(TCP se)100.8 287.2 Q(gment size is a function of the byte drop rate.)
-.165 E(In particular)5.5 E 2.75(,t)-.44 G(he optimal TCP)-2.75 E(se)
100.8 300.2 Q(gment size seems to be the one that k)-.165 E
(eeps the pack)-.11 E(et drop rate at most 15%, k)-.11 E(eeping the)-.11
E(TCP connection in the re)100.8 313.2 Q
(gime controlled by a 1/sqrt\(p\) sending rate, for pack)-.165 E
(et drop rate p.)-.11 E -.165(Fo)100.8 326.2 S 2.75(raT).165 G(CP pack)
-2.75 E(et size of S bytes \(including headers\), and a *byte* drop rat\
e of B, the)-.11 E(pack)100.8 339.2 Q(et drop rate is roughly B*S.)-.11
E -.165(Fo)5.5 G 2.75(rag).165 G -2.365 -.275(iv e)-2.75 H 2.75(nb).275
G(yte drop rate in this re)-2.75 E(gime, if the optimal)-.165 E
(TCP performance occurs with a pack)100.8 352.2 Q(et size chosen to gi)
-.11 E .33 -.165(ve a p)-.275 H(ack).165 E(et drop rate of at most)-.11
E(15%, k)100.8 365.2 Q(eeping the TCP connection out of the re)-.11 E
(gime of e)-.165 E(xponential back)-.165 E(of)-.11 E(fs of the)-.275 E
(retransmit timer)100.8 378.2 Q 2.75(,t)-.44 G
(hen this requires B*S = 0.15, or S = 0.15/B.)-2.75 E
(In the high-congestion re)100.8 394.8 Q(gime of T)-.165 E
(able 10, with high congestion and with byte drop rates)-.88 E
(of 0.01 and higher)100.8 407.8 Q 2.75(,t)-.44 G
(he TCP performance is dominated by the e)-2.75 E(xponential back)-.165
E(of)-.11 E 2.75(fo)-.275 G 2.75(ft)-2.75 G(he)-2.75 E
(retransmit timer re)100.8 420.8 Q -.055(ga)-.165 G(rdless of the se)
.055 E(gment size.)-.165 E(Ev)5.5 E(en a 40-byte pack)-.165 E
(et with a zero-byte)-.11 E(data se)100.8 433.8 Q(gment w)-.165 E
(ould ha)-.11 E .33 -.165(ve a p)-.22 H(ack).165 E
(et drop rate of at least 33%.)-.11 E(In this re)5.5 E
(gime, the optimal)-.165 E(TCP *sending* rate comes with a lar)100.8
446.8 Q(ge, 1460-byte data se)-.198 E(gment, b)-.165 E
(ut the TCP sending rate)-.22 E(is lo)100.8 459.8 Q 2.75(ww)-.275 G
(ith an)-2.75 E 2.75(ys)-.165 G -.165(eg)-2.75 G
(ment size, considerably less than one pack).165 E
(et per round-trip time.)-.11 E -.88(We)5.5 G(note that in this re)100.8
472.8 Q(gime, while a lar)-.165 E(ger pack)-.198 E(et gi)-.11 E -.165
(ve)-.275 G 2.75(sah).165 G(igher TCP *sending* rate, a smaller)-2.75 E
(pack)100.8 485.8 Q(et gi)-.11 E -.165(ve)-.275 G 2.75(sab).165 G
(etter *goodput* rate.)-2.75 E(In general, T)100.8 502.4 Q
(ables 8 and 9 sho)-.88 E 2.75(wg)-.275 G
(ood performance for TFRC-SP in en)-2.75 E(vironments with)-.44 E
(stable pack)100.8 515.4 Q
(et drop rates, where the decision to drop a pack)-.11 E
(et is independent of the pack)-.11 E(et)-.11 E 2.75(size. Ho)100.8
528.4 R(we)-.275 E -.165(ve)-.275 G .88 -.44(r, i).165 H 2.75(ns).44 G
(ome en)-2.75 E(vironments the pack)-.44 E(et size might af)-.11 E
(fect the lik)-.275 E(elihood that a)-.11 E(pack)100.8 541.4 Q
(et is dropped.)-.11 E -.165(Fo)5.5 G 2.75(re).165 G(xample, with hea)
-2.915 E(vy congestion and a Drop T)-.22 E(ail queue with a \214x)-.88 E
(ed)-.165 E(number of bytes rather than a \214x)100.8 554.4 Q
(ed number of pack)-.165 E(et-sized b)-.11 E(uf)-.22 E(fers, small pack)
-.275 E(ets might be)-.11 E(more lik)100.8 567.4 Q(ely than lar)-.11 E
(ge pack)-.198 E(ets to \214nd room at the end of an almost-full queue.)
-.11 E(As a)8.25 E(further complication, in a scenario with Acti)100.8
580.4 Q .33 -.165(ve Q)-.275 H(ueue Management, the A).165 E
(QM mechanism)-.605 E(could either be in pack)100.8 593.4 Q
(et mode, dropping each pack)-.11 E(et with equal probability)-.11 E
2.75(,o)-.715 G 2.75(ri)-2.75 G 2.75(nb)-2.75 G(yte)-2.75 E
(mode, dropping each byte with equal probability)100.8 606.4 Q 5.5(.S)
-.715 G(ections B.3 and B.4 sho)-5.5 E 2.75(ws)-.275 G(imulations)-2.75
E(with pack)100.8 619.4 Q(ets dropped at Drop T)-.11 E(ail or A)-.88 E
(QM queues, rather that from a probabilistic process.)-.605 E/F1 11
/Times-Bold@0 SF(B.3.)72 645.4 Q/F2 13/Times-Bold@0 SF -.13(Pa)5.5 G(ck)
.13 E(et Dr)-.13 E(opping Beha)-.234 E(vior at Routers with Dr)-.325 E
(op-T)-.234 E(ail Queues)-1.196 E F0
(One of the problems with comparing the throughput of tw)100.8 662 Q
2.75<6f8d>-.11 G -.275(ow)-2.75 G 2.75(su).275 G(sing dif)-2.75 E
(ferent pack)-.275 E(et)-.11 E(sizes is that the pack)100.8 675 Q
(et size itself can in\215uence the pack)-.11 E
(et drop rate [V00, WBL04].)-.11 E(Flo)72 733 Q(yd/K)-.11 E 268.584
(ohler Section)-.385 F 2.75(B.3. [P)2.75 F(age 28])-.165 E 0 Cg EP
%%Page: 29 29
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E(The def)
100.8 85 Q(ault TFRC w)-.11 E(as designed for rough f)-.11 E
(airness with TCP)-.11 E 2.75(,f)-1.221 G(or TFRC and TCP \215o)-2.75 E
(ws)-.275 E(with the same pack)100.8 98 Q(et size and e)-.11 E
(xperiencing the same pack)-.165 E(et drop rate.)-.11 E
(When the issue of)5.5 E -.11(fa)100.8 111 S(irness between \215o).11 E
(ws with dif)-.275 E(ferent pack)-.275 E
(ets sizes is addressed, it matters whether the)-.11 E(pack)100.8 124 Q
(et drop rates e)-.11 E(xperienced by the \215o)-.165 E
(ws is related to the pack)-.275 E(et size.)-.11 E(That is, are small)
5.5 E(pack)100.8 137 Q(ets just as lik)-.11 E(ely to be dropped as lar)
-.11 E(ge TCP pack)-.198 E(ets, or are the smaller pack)-.11 E(ets less)
-.11 E(lik)100.8 150 Q(ely to be dropped [WBL04].)-.11 E
(And what is the relationship between the pack)5.5 E(et-dropping)-.11 E
(beha)100.8 163 Q(vior of the path, and the loss e)-.22 E -.165(ve)-.275
G(nt measurements of TFRC?).165 E/F1 11/Courier@0 SF 6.6(<-----S)213 182
S(end Rates in Kbps - - - - >)-6.6 E 46.2(Web TCP)153.6 195 R
(\(1460B seg\))6.6 E(TFRC-SP \(200B seg\))33 E 13.2(Sessions DropRate)
140.4 208 R 19.8(SendRate DropRate)13.2 F(SendRate)13.2 E 13.2
(-------- --------)140.4 221 R 19.8(-------- --------)13.2 F(--------)
13.2 E 39.6(10 0.04)166.8 234 R 39.6(316.18 0.05)33 F(183.05)33 E 39.6
(25 0.07)166.8 247 R 39.6(227.47 0.07)33 F(181.23)33 E 39.6(50 0.08)
166.8 260 R 39.6(181.10 0.08)33 F(178.32)33 E 39.6(100 0.14)160.2 273 R
39.6(85.97 0.12)39.6 F(151.42)33 E 39.6(200 0.17)160.2 286 R 39.6
(61.20 0.14)39.6 F(73.88)39.6 E 39.6(400 0.20)160.2 299 R 39.6
(27.79 0.18)39.6 F(36.81)39.6 E 39.6(800 0.29 3.50 0.27)160.2 312 R
(16.33)39.6 E 39.6(1600 0.37 0.63 0.33 6.29)153.6 325 R
(Table 12: Drop and Send Rates for Drop-Tail Queues in Packets)127.2 351
Q F0 -.88(Ta)100.8 373.6 S(ble 12 sho).88 E(ws the results of the secon\
d half of 100-second simulations, with \214v)-.275 E 2.75(eT)-.165 G(CP)
-2.75 E(connections and \214v)100.8 386.6 Q 2.75(eT)-.165 G
(FRC-SP connections competing with web traf)-2.75 E
(\214c in a topology with)-.275 E 2.75(a3M)100.8 399.6 S
(bps shared link.)-2.75 E
(The TFRC-SP application generates 200-byte data pack)5.5 E(ets e)-.11 E
-.165(ve)-.275 G(ry 10).165 E(ms, for a maximum data rate of 160 Kbps.)
100.8 412.6 Q(The \214v)5.5 E 2.75(el)-.165 G(ong-li)-2.75 E -.165(ve)
-.275 G 2.75(dT).165 G(CP connections use a data)-2.75 E(pack)100.8
425.6 Q(et size of 1460 bytes, and the web traf)-.11 E
(\214c uses a data pack)-.275 E(et size of 512 bytes.)-.11 E(The \214v)
5.5 E(e)-.165 E(TCP connections ha)100.8 438.6 Q .33 -.165(ve r)-.22 H
(oundtrip times from 40 to 240 ms, and the \214v).165 E 2.75(eT)-.165 G
(FRC connections)-2.75 E(ha)100.8 451.6 Q .33 -.165(ve t)-.22 H
(he same set of round-trip times.).165 E(The SA)5.5 E
(CK TCP connections in these simulations)-.44 E(use the def)100.8 464.6
Q(ault parameters in the NS simulator)-.11 E 2.75(,w)-.44 G
(ith Limited T)-2.75 E(ransmit, and a minimum)-.385 E -1.628 -.66(RT O)
100.8 477.6 T(of 200 ms.)3.41 E
(Adding timestamps to the TCP connection didn')5.5 E 2.75(tc)-.198 G
(hange the results)-2.75 E(appreciably)100.8 490.6 Q 5.5(.T)-.715 G
(he simulations include re)-5.5 E -.165(ve)-.275 G(rse-path traf).165 E
(\214c, to add some small control)-.275 E(pack)100.8 503.6 Q
(ets to the forw)-.11 E(ard path, and some queueing delay to the re)-.11
E -.165(ve)-.275 G(rse path.).165 E(The number of)5.5 E
(web sessions is v)100.8 516.6 Q(aried to create dif)-.275 E(ferent le)
-.275 E -.165(ve)-.275 G(ls of congestion.).165 E(The Drop-T)5.5 E
(ail queue is in)-.88 E(units of pack)100.8 529.6 Q
(ets, which each pack)-.11 E(et arri)-.11 E
(ving to the queue requires a single b)-.275 E(uf)-.22 E(fer)-.275 E
2.75(,r)-.44 G -2.475 -.165(eg a)-2.75 H(rdless).165 E(of the pack)100.8
542.6 Q(et size.)-.11 E -.88(Ta)100.8 559.2 S(ble 12 sho).88 E(ws the a)
-.275 E -.165(ve)-.22 G(rage TCP and TFRC sending rates, each a).165 E
-.165(ve)-.22 G(raged o).165 E -.165(ve)-.165 G 2.75(rt).165 G(he \214v)
-2.75 E(e)-.165 E<8d6f>100.8 572.2 Q 2.75(ws. As)-.275 F -.165(ex)2.75 G
(pected, the TFRC-SP \215o).165 E(ws see similar pack)-.275 E
(et drop rates as the TCP \215o)-.11 E(ws,)-.275 E
(though the TFRC-SP \215o)100.8 585.2 Q(ws recei)-.275 E -.165(ve)-.275
G 2.75(sh).165 G(igher throughput than the TCP \215o)-2.75 E
(ws with pack)-.275 E(et drop)-.11 E(rates of 25% or higher)100.8 598.2
Q(.)-.605 E(Flo)72 733 Q(yd/K)-.11 E 268.584(ohler Section)-.385 F 2.75
(B.3. [P)2.75 F(age 29])-.165 E 0 Cg EP
%%Page: 30 30
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E/F1 11
/Courier@0 SF 6.6(<-----S)226.2 85 S(end Rates in Kbps - - - - >)-6.6 E
39.6(Web TCP)166.8 98 R(\(1460B seg\))6.6 E(TFRC-SP \(200B seg\))39.6 E
13.2(Sessions DropRate)153.6 111 R 19.8(SendRate DropRate)13.2 F
(SendRate)13.2 E 13.2(-------- --------)153.6 124 R 19.8
(-------- --------)13.2 F(--------)13.2 E 33(10 0.061)180 137 R 33
(239.81 0.004)33 F(185.19)33 E 33(25 0.089)180 150 R 33(189.02 0.006)33
F(184.95)33 E 33(50 0.141 99.46 0.013)180 163 R(185.07)33 E 33
(100 0.196 16.42 0.022)173.4 176 R(183.77)33 E 33(200 0.256)173.4 189 R
33(4.46 0.032)46.2 F(181.98)33 E 33(400 0.291)173.4 202 R 33(4.61 0.051)
46.2 F(151.88)33 E 33(800 0.487)173.4 215 R 33(1.01 0.078)46.2 F(113.10)
33 E 33(1600 0.648)166.8 228 R 33(0.67 0.121 65.17)46.2 F
(Table 13: Drop and Send Rates for Drop-Tail Queues in Bytes I:)107.4
267 Q(1460B TCP Segments)272.4 280 Q F0(Ho)100.8 302.6 Q(we)-.275 E
-.165(ve)-.275 G .88 -.44(r, t).165 H(he f).44 E
(airness results can change signi\214cantly if the Drop-T)-.11 E
(ail queue at the)-.88 E
(congested output link is in units of bytes rather than pack)100.8 315.6
Q 2.75(ets. F)-.11 F(or a queue in pack)-.165 E(ets, the)-.11 E
(queue has a \214x)100.8 328.6 Q(ed number of b)-.165 E(uf)-.22 E
(fers, and each b)-.275 E(uf)-.22 E(fer can hold e)-.275 E
(xactly one pack)-.165 E(et,)-.11 E(re)100.8 341.6 Q -.055(ga)-.165 G
(rdless of its size in bytes.).055 E -.165(Fo)5.5 G 2.75(raq).165 G
(ueue in bytes, the queue has a \214x)-2.75 E(ed number of)-.165 E
(*bytes*, and an almost-full queue might ha)100.8 354.6 Q .33 -.165
(ve r)-.22 H(oom for a small pack).165 E(et b)-.11 E(ut not for a lar)
-.22 E(ge)-.198 E 2.75(one. This,)100.8 367.6 R
(for a simulation with a Drop-T)2.75 E(ail queue in bytes, lar)-.88 E
(ge pack)-.198 E(ets are more lik)-.11 E(ely)-.11 E
(to be dropped than are small ones.)100.8 380.6 Q
(The NS simulator doesn')5.5 E 2.75(ty)-.198 G(et ha)-2.75 E .33 -.165
(ve a m)-.22 H(ore realistic).165 E
(intermediate model, where the queue has a \214x)100.8 393.6 Q
(ed number of b)-.165 E(uf)-.22 E(fers, each b)-.275 E(uf)-.22 E
(fer has a \214x)-.275 E(ed)-.165 E(number of bytes, and each pack)100.8
406.6 Q(et w)-.11 E(ould require one or more free b)-.11 E(uf)-.22 E
2.75(fers. In)-.275 F(this model, a)2.75 E(small pack)100.8 419.6 Q
(et w)-.11 E(ould use one b)-.11 E(uf)-.22 E(fer)-.275 E 2.75(,w)-.44 G
(hile a lar)-2.75 E(ger pack)-.198 E(et w)-.11 E(ould require se)-.11 E
-.165(ve)-.275 G(ral b).165 E(uf)-.22 E(fers.)-.275 E
(The scenarios in T)100.8 436.2 Q(able 13 are identical to those in T)
-.88 E(able 12, e)-.88 E(xcept that the Drop-T)-.165 E(ail queue)-.88 E
(is in units of bytes instead of pack)100.8 449.2 Q 2.75(ets. Thus,)-.11
F<8c76>2.75 E 2.75(eT)-.165 G(CP connections and \214v)-2.75 E 2.75(eT)
-.165 G(FRC-SP)-2.75 E(connections compete with web traf)100.8 462.2 Q
(\214c in a topology with a 3 Mbps shared link, with each)-.275 E
(TFRC-SP application generating 200-byte data pack)100.8 475.2 Q(ets e)
-.11 E -.165(ve)-.275 G(ry 10 ms, for a maximum data).165 E
(rate of 160 Kbps.)100.8 488.2 Q(The number of web sessions is v)5.5 E
(aried to create dif)-.275 E(ferent le)-.275 E -.165(ve)-.275 G(ls of)
.165 E 2.75(congestion. Ho)100.8 501.2 R(we)-.275 E -.165(ve)-.275 G .88
-.44(r, i).165 H(nstead of Drop-T).44 E
(ail queues able to accommodate at most a hundred)-.88 E(pack)100.8
514.2 Q(ets, the routers' Drop-T)-.11 E
(ail queues are each able to accommodate at most 50,000 bytes)-.88 E
(\(computed in NS as a hundred pack)100.8 527.2 Q
(ets times the mean pack)-.11 E(etsize of 500 bytes\).)-.11 E(As T)100.8
543.8 Q(able 13 sho)-.88 E(ws, with a Drop-T)-.275 E
(ail queue in bytes, the TFRC-SP \215o)-.88 E 2.75(ws)-.275 G
(ees a much)-2.75 E(smaller pack)100.8 556.8 Q
(et drop rate than the TCP \215o)-.11 E 1.43 -.715(w, a)-.275 H
(nd as a consequence recei).715 E -.165(ve)-.275 G 2.75(sam).165 G
(uch lar)-2.75 E(ger)-.198 E(sending rate.)100.8 569.8 Q -.165(Fo)5.5 G
2.75(rt).165 G(he simulations in T)-2.75 E(able 13, the TFRC-SP \215o)
-.88 E(ws use 200-byte data)-.275 E(se)100.8 582.8 Q
(gments, while the long-li)-.165 E -.165(ve)-.275 G 2.75(dT).165 G
(CP \215o)-2.75 E(ws use 1460-byte data se)-.275 E 2.75(gments. F)-.165
F(or e)-.165 E(xample,)-.165 E(when the \214v)100.8 595.8 Q 2.75(eT)
-.165 G(CP \215o)-2.75 E(ws and \214v)-.275 E 2.75(eT)-.165 G
(FRC-SP \215o)-2.75 E(ws share the link with 800 web sessions, the)-.275
E<8c76>100.8 608.8 Q 2.75(eT)-.165 G(CP \215o)-2.75 E(ws see an a)-.275
E -.165(ve)-.22 G
(rage drop rate of 49% in the second half of the simulation, while).165
E(the \214v)100.8 621.8 Q 2.75(eT)-.165 G(FRC-SP \215o)-2.75 E(ws recei)
-.275 E .33 -.165(ve a)-.275 H 2.75(na).165 G -.165(ve)-2.97 G
(rage drop rate of 8%, and as a consequence recei).165 E -.165(ve)-.275
G(more than 100 times the throughput of the TCP \215o)100.8 634.8 Q 2.75
(ws. This)-.275 F(raises serious questions about)2.75 E
(making the assumption that \215o)100.8 647.8 Q(ws with small pack)-.275
E(ets see the same pack)-.11 E(et drop rate as \215o)-.11 E(ws)-.275 E
(with lar)100.8 660.8 Q(ger pack)-.198 E 2.75(ets. Further)-.11 F -.11
(wo)2.75 G(rk will ha).11 E .33 -.165(ve t)-.22 H 2.75(oi).165 G
(nclude an in)-2.75 E -.165(ve)-.44 G(stig).165 E
(ation into the range of)-.055 E
(realistic Internet scenarios, in terms of whether lar)100.8 673.8 Q
(ge pack)-.198 E(ets are considerably more lik)-.11 E(ely)-.11 E
(to be dropped than are small ones.)100.8 686.8 Q(Flo)72 733 Q(yd/K)-.11
E 268.584(ohler Section)-.385 F 2.75(B.3. [P)2.75 F(age 30])-.165 E 0 Cg
EP
%%Page: 31 31
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E/F1 11
/Courier@0 SF 6.6(<-----S)226.2 85 S(end Rates in Kbps - - - - >)-6.6 E
46.2(Web TCP)166.8 98 R(\(512B seg\))6.6 E(TFRC-SP \(200B seg\))39.6 E
13.2(Sessions DropRate)153.6 111 R 19.8(SendRate DropRate)13.2 F
(SendRate)13.2 E 13.2(-------- --------)153.6 124 R 19.8
(-------- --------)13.2 F(--------)13.2 E 39.6(10 0.02)180 137 R 39.6
(335.05 0.00)33 F(185.16)33 E 39.6(25 0.02)180 150 R 39.6(289.94 0.00)33
F(185.36)33 E 39.6(50 0.04)180 163 R 39.6(139.99 0.01)33 F(184.98)33 E
39.6(100 0.06)173.4 176 R 39.6(53.50 0.01)39.6 F(184.66)33 E 39.6
(200 0.10)173.4 189 R 39.6(16.14 0.04)39.6 F(167.87)33 E 39.6
(400 0.16 6.36 0.07)173.4 202 R(114.85)33 E 39.6(800 0.24 0.90 0.11)
173.4 215 R(67.23)39.6 E 39.6(1600 0.42 0.35 0.18)166.8 228 R(39.32)39.6
E(Table 14: Drop and Send Rates for Drop-Tail Queues in Bytes II:)107.4
254 Q(512B TCP Segments)279 267 Q F0 -.88(Ta)100.8 289.6 S(ble 14 sho)
.88 E(ws that in the scenario the long-li)-.275 E -.165(ve)-.275 G 2.75
(dT).165 G(CP \215o)-2.75 E(ws recei)-.275 E .33 -.165(ve a h)-.275 H
(igher pack).165 E(et drop)-.11 E(rate than the TFRC-SP \215o)100.8
302.6 Q(ws, and recei)-.275 E .33 -.165(ve c)-.275 H
(onsiderably less throughput, e).165 E -.165(ve)-.275 G 2.75(nw).165 G
(hen the)-2.75 E(long-li)100.8 315.6 Q -.165(ve)-.275 G 2.75(dT).165 G
(CP \215o)-2.75 E(ws use 512-byte se)-.275 E(gments.)-.165 E 1.76 -.88
(To s)100.8 332.2 T(ho).88 E 2.75(wt)-.275 G(he potential ne)-2.75 E
-.055(ga)-.165 G(ti).055 E .33 -.165(ve e)-.275 H -.275(ff).165 G
(ect of TFRC-SP in such an en).275 E(vironment, we consider a)-.44 E
(simulation with N TCP \215o)100.8 345.2 Q(ws, N TFRC-SP \215o)-.275 E
(ws, and 10*N web sessions, for N ranging)-.275 E
(from 1 to 50, so that the demand increases from all types of traf)100.8
358.2 Q(\214c, with routers with Drop-)-.275 E -.88(Ta)100.8 371.2 S
(il queues in bytes.).88 E F1 6.6(<-----S)226.2 390.2 S
(end Rates in Kbps - - - - >)-6.6 E 39.6(Web TCP)166.8 403.2 R
(\(1460B seg\))6.6 E(TFRC-SP \(200B seg\))39.6 E 13.2(Sessions DropRate)
153.6 416.2 R 19.8(SendRate DropRate)13.2 F(SendRate)13.2 E 13.2
(-------- --------)153.6 429.2 R 19.8(-------- --------)13.2 F(--------)
13.2 E 33(10 0.014)180 442.2 R 33(2085.36 0.001)26.4 F(180.29)33 E 33
(20 0.040)180 455.2 R 33(788.88 0.004)33 F(183.87)33 E 33(30 0.074)180
468.2 R 33(248.80 0.006)33 F(182.94)33 E 33(40 0.113)180 481.2 R 33
(163.20 0.008)33 F(185.33)33 E 33(50 0.127 94.70 0.011)180 494.2 R
(185.14)33 E 33(60 0.177 53.24 0.015)180 507.2 R(185.30)33 E 33
(70 0.174 35.31 0.016)180 520.2 R(185.07)33 E 33(80 0.221 19.38 0.019)
180 533.2 R(183.91)33 E 33(90 0.188 15.63 0.019)180 546.2 R(180.42)33 E
33(100 0.265)173.4 559.2 R 33(7.08 0.023)46.2 F(176.71)33 E 33
(200 0.324)173.4 572.2 R 33(0.38 0.042)46.2 F(139.48)33 E 33(300 0.397)
173.4 585.2 R 33(0.32 0.076 93.53)46.2 F 33(400 0.529)173.4 598.2 R 33
(0.40 0.100 70.40)46.2 F 33(500 0.610)173.4 611.2 R 33(0.37 0.121 57.59)
46.2 F(Table 15: Drop and Send Rates for Drop-Tail Queues in Bytes III:)
107.4 637.2 Q(TFRC-SP, 1460B TCP Segments)219.6 650.2 Q F0(Flo)72 733 Q
(yd/K)-.11 E 268.584(ohler Section)-.385 F 2.75(B.3. [P)2.75 F(age 31])
-.165 E 0 Cg EP
%%Page: 32 32
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E/F1 11
/Courier@0 SF 6.6(<-----S)226.2 85 S(end Rates in Kbps - - - - >)-6.6 E
39.6(Web TCP)166.8 98 R(\(1460B seg\))6.6 E(TFRC \(200B seg\))46.2 E
13.2(Sessions DropRate)153.6 111 R 19.8(SendRate DropRate)13.2 F
(SendRate)13.2 E 13.2(-------- --------)153.6 124 R 19.8
(-------- --------)13.2 F(--------)13.2 E 33(10 0.016)180 137 R 33
(1926.00 0.002)26.4 F(178.47)33 E 33(20 0.030)180 150 R 33(805.20 0.003)
33 F(178.23)33 E 33(30 0.062)180 163 R 33(346.24 0.005)33 F(161.19)33 E
33(40 0.093)180 176 R 33(219.18 0.007)33 F(136.28)33 E 33(50 0.110)180
189 R 33(132.77 0.010 83.02)33 F 33(60 0.170 88.88 0.014 69.75)180 202 R
33(70 0.149 70.73 0.015 55.59)180 215 R 33(80 0.213 43.17 0.020 42.82)
180 228 R 33(90 0.188 36.19 0.017 43.61)180 241 R 33
(100 0.233 24.77 0.026 35.17)173.4 254 R 33(200 0.311)173.4 267 R 33
(5.46 0.034 24.85)46.2 F 33(300 0.367)173.4 280 R 33(3.74 0.049 20.19)
46.2 F 33(400 0.421)173.4 293 R 33(2.59 0.055 17.71)46.2 F 33(500 0.459)
173.4 306 R 33(1.10 0.069 15.95)46.2 F
(Table 16: Drop and Send Rates for Drop-Tail Queues in Bytes IV:)107.4
319 Q(Standard TFRC, 1460B TCP Segments)199.8 332 Q F0 -.88(Ta)100.8
354.6 S(ble 15 sho).88 E(ws simulations using TFRC-SP)-.275 E 2.75(,w)
-1.221 G(hile T)-2.75 E(able 16 sho)-.88 E(ws simulations using)-.275 E
(TFRC instead of TFRC-SP)100.8 367.6 Q 5.5(.T)-1.221 G
(his is the only dif)-5.5 E(ference between the simulations in the tw)
-.275 E(o)-.11 E 2.75(tables. Note)100.8 380.6 R
(that when TFRC-SP is used, the TCP \215o)2.75 E(ws and web traf)-.275 E
(\214c are essentially)-.275 E(starv)100.8 393.6 Q
(ed, while the TFRC-SP \215o)-.165 E(ws each continue to send 57 Kbps.)
-.275 E(In contrast, when)5.5 E
(standard TFRC is used instead of TFRC-SP for the \215o)100.8 406.6 Q
(ws sending 200-byte se)-.275 E(gments, the)-.165 E(TCP \215o)100.8
419.6 Q(ws are not starv)-.275 E(ed \(although the)-.165 E 2.75(ys)-.165
G(till don')-2.75 E 2.75(tr)-.198 G(ecei)-2.75 E .33 -.165(ve t)-.275 H
(heir "share" of the link).165 E(bandwidth when their pack)100.8 432.6 Q
(et drop rates are 30% or higher)-.11 E 2.75(.\) These)-.605 F(tw)2.75 E
2.75(os)-.11 G(ets of simulations)-2.75 E
(illustrate the dangers of a widespread deplo)100.8 445.6 Q
(yment of TFRC-SP in an en)-.11 E(vironment where)-.44 E(small pack)
100.8 458.6 Q(ets are less lik)-.11 E(ely to be dropped than lar)-.11 E
(ge ones.)-.198 E/F2 11/Times-Bold@0 SF(B.4.)72 484.6 Q/F3 13
/Times-Bold@0 SF -.13(Pa)5.5 G(ck).13 E(et Dr)-.13 E(opping Beha)-.234 E
(vior at Routers with A)-.325 E(QM)-.585 E F0(As e)100.8 501.2 Q
(xpected, the pack)-.165 E(et dropping beha)-.11 E(vior also can be v)
-.22 E(aried by v)-.275 E(arying the Acti)-.275 E .33 -.165(ve Q)-.275 H
(ueue).165 E(Management \(A)100.8 514.2 Q(QM\) mechanism in the router)
-.605 E 5.5(.W)-.605 G(hen the routers use RED \(Random Early)-5.5 E
(Detection\), there are se)100.8 527.2 Q -.165(ve)-.275 G
(ral parameters than can af).165 E(fect the pack)-.275 E
(et dropping or marking)-.11 E(beha)100.8 540.2 Q
(vior as a function of the pack)-.22 E(et size.)-.11 E
(First, as with Drop-T)100.8 556.8 Q
(ail, the RED queue can be either in units of pack)-.88 E
(ets or of bytes.)-.11 E(This)5.5 E(can af)100.8 569.8 Q(fect the pack)
-.275 E(et dropping beha)-.11 E
(vior when RED is unable to control the a)-.22 E -.165(ve)-.22 G
(rage queue).165 E(size, and the queue o)100.8 582.8 Q -.165(ve)-.165 G
(r\215o).165 E(ws.)-.275 E(Second, and orthogonally)100.8 599.4 Q 2.75
(,R)-.715 G(ED can be con\214gured to be either in pack)-2.75 E
(et mode or in byte)-.11 E 2.75(mode. In)100.8 612.4 R(pack)2.75 E
(et mode, each *pack)-.11 E
(et* has the same probability of being dropped by RED,)-.11 E(while in \
byte mode, each *byte* has the same probability of being dropped.)100.8
625.4 Q(In pack)5.5 E(et)-.11 E(mode, lar)100.8 638.4 Q(ge-pack)-.198 E
(et and small-pack)-.11 E(et \215o)-.11 E(ws recei)-.275 E .33 -.165
(ve r)-.275 H(oughly the same pack).165 E(et drop rate,)-.11 E
(while in byte mode, lar)100.8 651.4 Q(ge-pack)-.198 E
(et and small-pack)-.11 E(et \215o)-.11 E
(ws with the same throughput in bps)-.275 E(recei)100.8 664.4 Q .33
-.165(ve r)-.275 H(oughly the same *number* of pack).165 E(et drops.)
-.11 E([EA03] assessed the impact of byte)5.5 E(vs. pack)100.8 677.4 Q
(et modes on RED performance.)-.11 E(Flo)72 733 Q(yd/K)-.11 E 268.584
(ohler Section)-.385 F 2.75(B.4. [P)2.75 F(age 32])-.165 E 0 Cg EP
%%Page: 33 33
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E
(The simulations reported belo)100.8 85 Q 2.75(ws)-.275 G(ho)-2.75 E
2.75(wt)-.275 G(hat for RED in pack)-2.75 E(et mode, the pack)-.11 E
(et drop rates)-.11 E(for the TFRC-SP \215o)100.8 98 Q
(ws are similar to those for the TCP \215o)-.275 E
(ws, with a resulting acceptable)-.275 E
(throughput for the TFRC-SP \215o)100.8 111 Q 5.5(ws. This)-.275 F
(is true with the queue in pack)2.75 E(ets or in bytes, and)-.11 E
(with or without Adapti)100.8 124 Q .33 -.165(ve R)-.275 H
(ED \(discussed belo).165 E 2.75(w\). As)-.275 F(we sho)2.75 E 2.75(wb)
-.275 G(elo)-2.75 E 1.43 -.715(w, t)-.275 H(his f).715 E(airness)-.11 E
(between TCP and TFRC-SP \215o)100.8 137 Q
(ws does not hold for RED in byte mode.)-.275 E
(The third RED parameter that af)100.8 153.6 Q(fects the pack)-.275 E
(et dropping or marking beha)-.11 E(vior as a)-.22 E(function of pack)
100.8 166.6 Q
(et size is whether the RED mechanism is using Standard RED or Adapti)
-.11 E -.165(ve)-.275 G 2.75(RED; Adapti)100.8 179.6 R .33 -.165(ve R)
-.275 H(ED tries to maintain the same a).165 E -.165(ve)-.22 G
(rage queue size, re).165 E -.055(ga)-.165 G(rdless of the).055 E(pack)
100.8 192.6 Q(et drop rate.)-.11 E(The use of Adapti)5.5 E .33 -.165
(ve R)-.275 H(ED allo).165 E(ws the RED mechanism to function more)-.275
E(ef)100.8 205.6 Q(fecti)-.275 E -.165(ve)-.275 G
(ly in the presence of high pack).165 E
(et drop rates \(e.g., greater than 10%\).)-.11 E -.44(Wi)5.5 G(thout)
.44 E(Adapti)100.8 218.6 Q .33 -.165(ve R)-.275 H(ED, there is a \214x)
.165 E(ed dropping threshold, and all arri)-.165 E(ving pack)-.275 E
(ets are dropped)-.11 E(when the dropping or marking rate e)100.8 231.6
Q(xceeds this threshold.)-.165 E(In contrast, with Adapti)5.5 E .33
-.165(ve R)-.275 H(ED,).165 E
(the dropping function is adapted to accommodate high-drop-rate re)100.8
244.6 Q 2.75(gimes. One)-.165 F
(consequence is that when byte mode is used with Adapti)100.8 257.6 Q
.33 -.165(ve R)-.275 H(ED, the byte mode e).165 E(xtends)-.165 E -2.365
-.275(ev e)100.8 270.6 T 2.75(nt).275 G 2.75(oh)-2.75 G
(igh-drop-rate re)-2.75 E 2.75(gimes. When)-.165 F
(byte mode is used with standard RED, ho)2.75 E(we)-.275 E -.165(ve)
-.275 G .88 -.44(r, t).165 H(he).44 E
(byte mode is no longer in use when the drop rate e)100.8 283.6 Q
(xceeds the \214x)-.165 E(ed dropping threshold \(set)-.165 E(by def)
100.8 296.6 Q(ault to 10% in the NS simulator\).)-.11 E
(In the simulations in this section, we e)100.8 313.2 Q
(xplore the TFRC-SP beha)-.165 E(vior o)-.22 E -.165(ve)-.165 G 2.75(rs)
.165 G(ome of this)-2.75 E(range of scenarios.)100.8 326.2 Q
(In this simulations, as in Section B.3 abo)5.5 E -.165(ve)-.165 G 2.75
(,t).165 G(he application for the)-2.75 E(TFRC-SP \215o)100.8 339.2 Q
2.75(wu)-.275 G(ses 200-byte data pack)-2.75 E(ets, generating 100 pack)
-.11 E(ets per second.)-.11 E/F1 11/Courier@0 SF 6.6(<-----S)226.2 358.2
S(end Rates in Kbps - - - - >)-6.6 E 46.2(Web TCP)166.8 371.2 R
(\(1460B seg\))6.6 E(TFRC-SP \(200B seg\))33 E 13.2(Sessions DropRate)
153.6 384.2 R 19.8(SendRate DropRate)13.2 F(SendRate)13.2 E 13.2
(-------- --------)153.6 397.2 R 19.8(-------- --------)13.2 F(--------)
13.2 E 39.6(10 0.05)180 410.2 R 39.6(305.76 0.04)33 F(182.82)33 E 39.6
(25 0.06)180 423.2 R 39.6(224.16 0.06)33 F(175.91)33 E 39.6(50 0.09)180
436.2 R 39.6(159.12 0.08)33 F(152.51)33 E 39.6(100 0.13)173.4 449.2 R
39.6(90.77 0.11)39.6 F(106.13)33 E 39.6(200 0.14)173.4 462.2 R 39.6
(48.53 0.14)39.6 F(70.25)39.6 E 39.6(400 0.20)173.4 475.2 R 39.6
(22.08 0.19)39.6 F(41.50)39.6 E 39.6(800 0.27 3.55 0.25)173.4 488.2 R
(17.50)39.6 E 39.6(1600 0.42 1.87 0.34 8.81)166.8 501.2 R
(Table 17: Drop and Send Rates for RED Queues in Packet Mode)114 527.2 Q
F0 -.165(Fo)100.8 549.8 S 2.75(rt).165 G(he simulations in T)-2.75 E
(able 17, with a congested router with a RED queue in pack)-.88 E
(et mode,)-.11 E(the results are similar to those with a Drop-T)100.8
562.8 Q(ail queue in pack)-.88 E(ets, as in T)-.11 E(able 12 abo)-.88 E
-.165(ve)-.165 G(.).165 E(The TFRC-SP \215o)100.8 575.8 Q 2.75(wr)-.275
G(ecei)-2.75 E -.165(ve)-.275 G 2.75(ss).165 G(imilar pack)-2.75 E
(et drop rates as the TCP \215o)-.11 E 1.43 -.715(w, t)-.275 H
(hough it recei).715 E -.165(ve)-.275 G(s).165 E
(higher throughput in the more congested en)100.8 588.8 Q 2.75
(vironments. The)-.44 F(simulations are similar with a)2.75 E
(RED queue in pack)100.8 601.8 Q
(et mode with the queue in bytes, and with or without Adapti)-.11 E .33
-.165(ve R)-.275 H 2.75(ED. In).165 F(these simulations, TFRC-SP gi)
100.8 614.8 Q -.165(ve)-.275 G 2.75(sr).165 G
(oughly the desired performance.)-2.75 E(Flo)72 733 Q(yd/K)-.11 E
268.584(ohler Section)-.385 F 2.75(B.4. [P)2.75 F(age 33])-.165 E 0 Cg
EP
%%Page: 34 34
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E/F1 11
/Courier@0 SF 6.6(<-----S)226.2 85 S(end Rates in Kbps - - - - >)-6.6 E
39.6(Web TCP)166.8 98 R(\(1460B seg\))6.6 E(TFRC-SP \(200B seg\))39.6 E
13.2(Sessions DropRate)153.6 111 R 19.8(SendRate DropRate)13.2 F
(SendRate)13.2 E 13.2(-------- --------)153.6 124 R 19.8
(-------- --------)13.2 F(--------)13.2 E 39.6(10 0.06)180 137 R 39.6
(272.16 0.02)33 F(184.37)33 E 39.6(25 0.07)180 150 R 39.6(175.82 0.02)33
F(184.06)33 E 39.6(50 0.10)180 163 R 39.6(75.65 0.04)39.6 F(180.56)33 E
39.6(100 0.14)173.4 176 R 39.6(38.98 0.07)39.6 F(151.65)33 E 39.6
(200 0.19)173.4 189 R 39.6(16.66 0.11)39.6 F(106.80)33 E 39.6
(400 0.26 4.85 0.15)173.4 202 R(69.41)39.6 E 39.6(800 0.35 3.12 0.20)
173.4 215 R(27.07)39.6 E 39.6(1600 0.42 0.67 0.29)166.8 228 R(10.68)39.6
E(Table 18: Drop and Send Rates for RED Queues in Byte Mode)127.2 254 Q
F0 -.88(Ta)100.8 276.6 S(ble 18 sho).88 E
(ws that with a standard RED queue in byte mode instead of pack)-.275 E
(et mode, there)-.11 E(is a some)100.8 289.6 Q(what greater dif)-.275 E
(ferent between the pack)-.275 E
(et drop rates between the TCP and TFRC-)-.11 E(SP \215o)100.8 302.6 Q
(ws, particularly for lo)-.275 E(wer pack)-.275 E(et drop rates.)-.11 E
-.165(Fo)5.5 G 2.75(rt).165 G(he simulation in T)-2.75 E(able 18, the)
-.88 E(pack)100.8 315.6 Q(et drop rates for the TCP \215o)-.11 E
(ws can range from 1 1/2 to four times greater than the)-.275 E(pack)
100.8 328.6 Q(et drop rates for the TFRC-SP \215o)-.11 E 2.75(ws. Ho)
-.275 F(we)-.275 E -.165(ve)-.275 G .88 -.44(r, b).165 H
(ecause the TFRC-SP \215o).44 E 2.75(wh)-.275 G(as an)-2.75 E
(upper bound on the sending rate, its sending rate is not af)100.8 341.6
Q(fected in the lo)-.275 E(wer pack)-.275 E(et-drop-)-.11 E(rate re)
100.8 354.6 Q(gimes; its sending rate is only af)-.165 E
(fected in the re)-.275 E(gimes with pack)-.165 E(et drop rates of 10%)
-.11 E(or more.)100.8 367.6 Q
(The sending rate for TFRC-SP in the scenarios in T)5.5 E
(able 18 with higher pack)-.88 E(et)-.11 E(drop rates are greater than \
desired, e.g., for the scenarios with 400 web sessions or greater)100.8
380.6 Q(.)-.605 E(Ho)100.8 393.6 Q(we)-.275 E -.165(ve)-.275 G .88 -.44
(r, t).165 H
(he results with TFRC-SP are at least better than that of small-pack).44
E(et \215o)-.11 E(ws with)-.275 E(no congestion control at all.)100.8
406.6 Q F1 6.6(<-----S)226.2 425.6 S(end Rates in Kbps - - - - >)-6.6 E
46.2(Web TCP)166.8 438.6 R(\(512B seg\))6.6 E(TFRC-SP \(200B seg\))39.6
E 13.2(Sessions DropRate)153.6 451.6 R 19.8(SendRate DropRate)13.2 F
(SendRate)13.2 E 13.2(-------- --------)153.6 464.6 R 19.8
(-------- --------)13.2 F(--------)13.2 E 39.6(10 0.01)180 477.6 R 39.6
(337.86 0.01)33 F(184.06)33 E 39.6(25 0.02)180 490.6 R 39.6(258.71 0.01)
33 F(184.03)33 E 39.6(50 0.02)180 503.6 R 39.6(184.71 0.01)33 F(183.99)
33 E 39.6(100 0.04)173.4 516.6 R 39.6(63.63 0.03)39.6 F(184.43)33 E 39.6
(200 0.08)173.4 529.6 R 39.6(28.95 0.06)39.6 F(149.80)33 E 39.6
(400 0.12)173.4 542.6 R 39.6(17.03 0.10)39.6 F(88.21)39.6 E 39.6
(800 0.24 5.94 0.15)173.4 555.6 R(36.80)39.6 E 39.6(1600 0.32 3.37 0.21)
166.8 568.6 R(19.45)39.6 E
(Table 19: Drop and Send Rates for RED Queues in Byte Mode)127.2 594.6 Q
F0 -.88(Ta)100.8 617.2 S(ble 19 sho).88 E
(ws that with a standard RED queue in byte mode and with long-li)-.275 E
-.165(ve)-.275 G 2.75(dT).165 G(CP)-2.75 E<8d6f>100.8 630.2 Q
(ws with 512-byte data se)-.275 E(gments, there is only a moderate dif)
-.165 E(ference between the pack)-.275 E(et)-.11 E
(drop rate for the 552-byte TCP pack)100.8 643.2 Q
(ets and the 240-byte TFRC-SP pack)-.11 E 2.75(ets. Ho)-.11 F(we)-.275 E
-.165(ve)-.275 G .88 -.44(r, t).165 H(he).44 E
(sending rate for TFRC-SP in the scenarios in T)100.8 656.2 Q
(able 19 with higher pack)-.88 E(et drop rates are)-.11 E
(still greater than desired, e)100.8 669.2 Q -.165(ve)-.275 G 2.75(ng)
.165 G -2.365 -.275(iv e)-2.75 H 2.75(nt).275 G(he goal of f)-2.75 E
(airness with TCP \215o)-.11 E(ws with 1500-byte)-.275 E(data se)100.8
682.2 Q(gments instead of 512-byte data se)-.165 E(gments.)-.165 E(Flo)
72 733 Q(yd/K)-.11 E 268.584(ohler Section)-.385 F 2.75(B.4. [P)2.75 F
(age 34])-.165 E 0 Cg EP
%%Page: 35 35
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E/F1 11
/Courier@0 SF 6.6(<-----S)226.2 85 S(end Rates in Kbps - - - - >)-6.6 E
39.6(Web TCP)166.8 98 R(\(1460B seg\))6.6 E(TFRC-SP \(200B seg\))39.6 E
13.2(Sessions DropRate)153.6 111 R 19.8(SendRate DropRate)13.2 F
(SendRate)13.2 E 13.2(-------- --------)153.6 124 R 19.8
(-------- --------)13.2 F(--------)13.2 E 39.6(10 0.04)180 137 R 39.6
(318.10 0.02)33 F(185.34)33 E 39.6(25 0.08)180 150 R 39.6(175.34 0.03)33
F(184.38)33 E 39.6(50 0.10)180 163 R 39.6(81.60 0.04)39.6 F(181.95)33 E
39.6(100 0.12)173.4 176 R 39.6(28.51 0.05)39.6 F(178.79)33 E 39.6
(200 0.20 3.65 0.06)173.4 189 R(173.78)33 E 39.6(400 0.27 1.44 0.08)
173.4 202 R(161.41)33 E 39.6(800 0.40 0.58 0.06)173.4 215 R(159.62)33 E
39.6(1600 0.55 0.29 0.02)166.8 228 R(180.92)33 E
(Table 20: Drop and Send Rates with Adaptive RED Queues in Byte Mode)
100.8 254 Q F0 -.165(Fo)100.8 276.6 S 2.75(rt).165 G
(he simulations in T)-2.75 E
(able 20, the congested router uses an Adapti)-.88 E .33 -.165(ve R)
-.275 H(ED queue in byte).165 E(mode.)100.8 289.6 Q -.165(Fo)100.8 306.2
S 2.75(rt).165 G(his router)-2.75 E 2.75(,t)-.44 G
(he output queue is in units of bytes rather than of pack)-2.75 E
(ets, each *byte* is)-.11 E(dropped with the same probability)100.8
319.2 Q 2.75(,a)-.715 G(nd because of the use of Adapti)-2.75 E .33
-.165(ve R)-.275 H(ED, this byte-).165 E(dropping mode e)100.8 332.2 Q
(xtends e)-.165 E -.165(ve)-.275 G 2.75(nf).165 G(or the high-pack)-2.75
E(et-drop-rate re)-.11 E(gime.)-.165 E(As T)100.8 348.8 Q(able 20 sho)
-.88 E(ws, for a scenario with Adapti)-.275 E .33 -.165(ve R)-.275 H
(ED in byte mode, the pack).165 E(et drop rate)-.11 E
(for the TFRC-SP traf)100.8 361.8 Q(\214c is *much* lo)-.275 E
(wer than that for the TCP traf)-.275 E(\214c, and as a)-.275 E
(consequence, the sending rate for the TFRC-SP traf)100.8 374.8 Q
(\214c in a highly congested en)-.275 E(vironment is)-.44 E
(*much* higher than that of the TCP traf)100.8 387.8 Q 2.75(\214c. In)
-.275 F -.11(fa)2.75 G(ct, in these scenarios the TFRC-SP).11 E
(congestion control mechanisms are lar)100.8 400.8 Q(gely inef)-.198 E
(fecti)-.275 E .33 -.165(ve f)-.275 H(or the small-pack).165 E(et traf)
-.11 E(\214c.)-.275 E(The simulation with 1600 web serv)100.8 417.4 Q
(ers is of particular concern, because the TCP pack)-.165 E(et)-.11 E
(drop rate increases, while the TFRC-SP pack)100.8 430.4 Q
(et drop rate decreases signi\214cantly)-.11 E 5.5(.T)-.715 G(his is)
-5.5 E(due to a detail of the Adapti)100.8 443.4 Q .33 -.165(ve R)-.275
H(ED implementation in the NS simulator).165 E 2.75(,a)-.44 G
(nd not about the)-2.75 E(dynamics of TFRC-SP)100.8 456.4 Q 5.5(.I)
-1.221 G 2.75(np)-5.5 G(articular)-2.75 E 2.75(,A)-.44 G(dapti)-2.75 E
.33 -.165(ve R)-.275 H(ED is con\214gured not to "adapt" to pack).165 E
(et)-.11 E(drop rates o)100.8 469.4 Q -.165(ve)-.165 G 2.75(r5).165 G
2.75(0%. When)-2.75 F(the pack)2.75 E
(et drop rate is at most 50%, Adapti)-.11 E .33 -.165(ve R)-.275 H
(ED is).165 E(moderately successful in k)100.8 482.4 Q(eeping the pack)
-.11 E(et drop rate proportional to the pack)-.11 E(et size - TCP)-.11 E
(pack)100.8 495.4 Q(ets are six times lar)-.11 E
(ger than the TFRC-SP pack)-.198 E
(ets \(including headers\), so the TCP)-.11 E(pack)100.8 508.4 Q
(ets should see a pack)-.11 E(et drop rate roughly six times lar)-.11 E
(ger)-.198 E 5.5(.B)-.605 G(ut for pack)-5.5 E(et drop rates)-.11 E
-.165(ove)100.8 521.4 S 2.75(r5).165 G(0%, Adapti)-2.75 E .33 -.165
(ve R)-.275 H(ED is no longer in this re).165 E
(gime, so it is no longer successful in)-.165 E -.11(ke)100.8 534.4 S
(eping the drop rate for TCP pack).11 E
(ets at most six times the drop rate of the TFRC-SP)-.11 E(pack)100.8
547.4 Q(ets.)-.11 E 1.76 -.88(We n)100.8 564 T(ote that the unf).88 E
(airness in these simulations, in f)-.11 E -.22(avo)-.11 G 2.75(ro).22 G
2.75(fT)-2.75 G(FRC-SP)-2.75 E 2.75(,i)-1.221 G 2.75(se)-2.75 G -.165
(ve)-3.025 G 2.75(nm).165 G(ore se)-2.75 E -.165(ve)-.275 G(re).165 E
(than the unf)100.8 577 Q(airness sho)-.11 E(wn in T)-.275 E
(able 13 for a Drop-T)-.88 E(ail queue in bytes.)-.88 E
(At the same time, it)5.5 E(is not kno)100.8 590 Q(wn if there is an)
-.275 E 2.75(yd)-.165 G(eplo)-2.75 E(yment in the Internet of an)-.11 E
2.75(yr)-.165 G(outers with Adapti)-2.75 E .33 -.165(ve R)-.275 H(ED in)
.165 E(byte mode, or of an)100.8 603 Q 2.75(yA)-.165 G
(QM mechanisms with similar beha)-3.355 E 2.75(vior; we)-.22 F(don')2.75
E 2.75(tk)-.198 G(no)-2.75 E 2.75(wt)-.275 G(he e)-2.75 E(xtent)-.165 E
(of the deplo)100.8 616 Q(yment of standard RED, or or an)-.11 E 2.75
(yo)-.165 G 2.75(ft)-2.75 G(he proposed A)-2.75 E(QM mechanisms.)-.605 E
(Flo)72 733 Q(yd/K)-.11 E 268.584(ohler Section)-.385 F 2.75(B.4. [P)
2.75 F(age 35])-.165 E 0 Cg EP
%%Page: 36 36
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E/F1 11
/Courier@0 SF 6.6(<-----S)226.2 85 S(end Rates in Kbps - - - - >)-6.6 E
46.2(Web TCP)166.8 98 R(\(512B seg\))6.6 E(TFRC-SP \(200B seg\))39.6 E
13.2(Sessions DropRate)153.6 111 R 19.8(SendRate DropRate)13.2 F
(SendRate)13.2 E 13.2(-------- --------)153.6 124 R 19.8
(-------- --------)13.2 F(--------)13.2 E 39.6(10 0.01)180 137 R 39.6
(306.56 0.01)33 F(185.11)33 E 39.6(25 0.02)180 150 R 39.6(261.41 0.01)33
F(184.41)33 E 39.6(50 0.02)180 163 R 39.6(185.07 0.01)33 F(184.54)33 E
39.6(100 0.04)173.4 176 R 39.6(59.25 0.03)39.6 F(181.58)33 E 39.6
(200 0.08)173.4 189 R 39.6(16.32 0.06)39.6 F(150.87)33 E 39.6(400 0.12)
173.4 202 R 39.6(11.53 0.10)39.6 F(98.10)39.6 E 39.6(800 0.25 5.85 0.15)
173.4 215 R(46.59)39.6 E 39.6(1600 0.32 1.43 0.22)166.8 228 R(19.40)39.6
E(Table 21: Drop and Send Rates for Adaptive RED Queues in Byte Mode)
100.8 254 Q F0 -.88(Ta)100.8 276.6 S(ble 21 sho).88 E
(ws that when TFRC-SP with 240-byte pack)-.275 E
(ets competes with TCP with)-.11 E(552-byte pack)100.8 289.6 Q
(ets in a scenario with Adapti)-.11 E .33 -.165(ve R)-.275 H
(ED in byte mode, the TFRC-SP \215o).165 E(ws still)-.275 E(recei)100.8
302.6 Q .33 -.165(ve m)-.275 H(ore bandwidth that the TCP \215o).165 E
(ws, b)-.275 E(ut the le)-.22 E -.165(ve)-.275 G 2.75(lo).165 G 2.75(fu)
-2.75 G(nf)-2.75 E(airness is less se)-.11 E -.165(ve)-.275 G(re, and)
.165 E(the pack)100.8 315.6 Q(et drop rates of the TCP \215o)-.11 E
(ws is at most twice that of the TFRC-SP \215o)-.275 E 2.75(ws. That)
-.275 F(is,)2.75 E(the TFRC-SP \215o)100.8 328.6 Q(ws still recei)-.275
E .33 -.165(ve m)-.275 H(ore than their share of the bandwidth, b).165 E
(ut the TFRC-SP)-.22 E(congestion control is more ef)100.8 341.6 Q
(fecti)-.275 E .33 -.165(ve t)-.275 H(hat than in T).165 E(able 20 abo)
-.88 E -.165(ve)-.165 G(.).165 E/F2 11/Times-Bold@0 SF(C.)72 367.6 Q/F3
14/Times-Bold@0 SF -.35(Ap)5.5 G(pendix: Exploring P).35 E
(ossible Oscillations in the Loss Ev)-.28 E(ent Rate)-.14 E F0
(TFRC-SP estimates the loss interv)100.8 384.2 Q(al size dif)-.275 E
(ferently for short and long loss interv)-.275 E(als,)-.275 E
(counting only one loss e)100.8 397.2 Q -.165(ve)-.275 G
(nt for long loss interv).165 E(als, b)-.275 E(ut counting all pack)-.22
E(et losses as loss)-.11 E -2.365 -.275(ev e)100.8 410.2 T
(nts for the short loss interv).275 E 2.75(als. One)-.275 F
(question that has been raised is whether this can)2.75 E
(lead to oscillations in the a)100.8 423.2 Q -.165(ve)-.22 G
(rage loss e).165 E -.165(ve)-.275 G(nt rate in en).165 E
(vironments where there are man)-.44 E(y)-.165 E(pack)100.8 436.2 Q
(et drops in a single loss e)-.11 E -.165(ve)-.275 G(nt, and loss e).165
E -.165(ve)-.275 G(nts switch from short to long and vice).165 E -.165
(ve)100.8 449.2 S 2.75(rsa. As).165 F(an e)2.75 E
(xample, consider a loss interv)-.165 E(al with N pack)-.275 E
(ets, including N/4 losses.)-.11 E(If this)5.5 E(loss interv)100.8 462.2
Q(al is short \(at most tw)-.275 E 2.75(or)-.11 G
(ound-trip times\), the loss interv)-2.75 E(al length is measured as)
-.275 E 2.75(4p)100.8 475.2 S(ack)-2.75 E 2.75(ets. If)-.11 F
(this loss interv)2.75 E(al is long, then the loss interv)-.275 E
(al length is measured as N)-.275 E(pack)100.8 488.2 Q(ets.)-.11 E
(If the loss interv)100.8 504.8 Q
(al switching from short to long and back leads to se)-.275 E -.165(ve)
-.275 G(re oscillations in the).165 E -2.475 -.22(av e)100.8 517.8 T
(rage pack).22 E(et drop rate, and therefore in the allo)-.11 E
(wed sending rate, one solution w)-.275 E(ould be)-.11 E(to ha)100.8
530.8 Q .33 -.165(ve a m)-.22 H
(ore gradual transition between the calculation of the loss interv).165
E(al length for)-.275 E(short and long loss interv)100.8 543.8 Q 2.75
(als. F)-.275 F(or e)-.165 E(xample, one possibility w)-.165 E
(ould be to use all of the pack)-.11 E(et)-.11 E
(losses for a loss interv)100.8 556.8 Q
(al of one round-trip time in calculating the loss interv)-.275 E
(al length, to use)-.275 E(2/3-rds of the pack)100.8 569.8 Q
(et losses from a loss interv)-.11 E(al of tw)-.275 E 2.75(or)-.11 G
(ound-trip times, to use 1/3-rd of the)-2.75 E(pack)100.8 582.8 Q
(et losses from a loss interv)-.11 E
(al of three round-trip time, and to use only one pack)-.275 E(et loss)
-.11 E(from a loss interv)100.8 595.8 Q
(al of four or more round-trip times.)-.275 E
(This more gradual mechanism w)5.5 E(ould)-.11 E(gi)100.8 608.8 Q .33
-.165(ve a g)-.275 H
(radual transition to counting all losses for a loss interv).165 E
(al of only one round-trip)-.275 E(time, and counting only one loss e)
100.8 621.8 Q -.165(ve)-.275 G(nt for a loss interv).165 E
(al of four or more round-trip times.)-.275 E(Ho)100.8 638.4 Q(we)-.275
E -.165(ve)-.275 G .88 -.44(r, o).165 H(ur simulations so f).44 E(ar ha)
-.11 E .33 -.165(ve n)-.22 H(ot sho).165 E(wn a great dif)-.275 E
(ference in oscillations in the)-.275 E(estimate loss e)100.8 651.4 Q
-.165(ve)-.275 G(nt rate between the def).165 E
(ault mechanism for estimating the loss interv)-.11 E(al)-.275 E
(length for short loss interv)100.8 664.4 Q
(al and the gradual mechanism described abo)-.275 E -.165(ve)-.165 G 5.5
(.S).165 G(imulation)-5.5 E(scripts are a)100.8 677.4 Q -.275(va)-.22 G
(ilable from [V).275 E 2.75(OIPSIMS]. Therefore,)-.44 F(for no)2.75 E
2.75(ww)-.275 G 2.75(ea)-2.75 G(re staying with the simple)-2.75 E(def)
100.8 690.4 Q(ault mechanism for estimating the loss e)-.11 E -.165(ve)
-.275 G(nt rate for short loss interv).165 E(als described in)-.275 E
(Flo)72 733 Q(yd/K)-.11 E 276.834(ohler Section)-.385 F 2.75(C. [P)2.75
F(age 36])-.165 E 0 Cg EP
%%Page: 37 37
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E
(this document.)100.8 85 Q/F1 11/Times-Bold@0 SF -.22(D.)72 111 S/F2 14
/Times-Bold@0 SF -.35(Ap)5.72 G(pendix: A Discussion of P).35 E(ack)-.14
E(et Size and P)-.14 E(ack)-.14 E(et Dr)-.14 E(opping)-.252 E F0
(The table belo)100.8 127.6 Q 2.75(wg)-.275 G -2.365 -.275(iv e)-2.75 H
2.75(sag).275 G(eneral summary of the relati)-2.75 E .33 -.165(ve a)
-.275 H(dv).165 E(antages of pack)-.275 E(et-dropping)-.11 E(beha)100.8
140.6 Q(vior at routers independent of pack)-.22 E(et size, v)-.11 E
(ersus pack)-.165 E(et dropping beha)-.11 E(vior where)-.22 E(lar)100.8
153.6 Q(ge pack)-.198 E(ets are more lik)-.11 E
(ely to be dropped than small ones.)-.11 E/F3 11/Courier@0 SF
(Advantages of Packet Dropping Independent of Packet Size:)100.8 172.6 Q
(---------------------------------------------------------)100.8 185.6 Q
6.6(1. Adds)100.8 198.6 R
(another incentive for end nodes to use large packets.)6.6 E 6.6
(2. Matches)100.8 224.6 R
(an environment with a limitation in pps rather than)6.6 E(bps.)127.2
237.6 Q(---------------------------------------------------------)100.8
250.6 Q(Advantages of Packet Dropping as a Function of Packet Size:)
100.8 276.6 Q(---------------------------------------------------------)
100.8 289.6 Q 6.6(1. Small)100.8 302.6 R
(control packets are less likely to be dropped than are)6.6 E
(large data packets, improving TCP performance.)127.2 315.6 Q 6.6
(2. Matches)100.8 341.6 R
(an environment with a limitation in bps rather than)6.6 E(pps.)127.2
354.6 Q 6.6(3. Reduces)100.8 380.6 R
(the penalty of TCP and other transport protocols)6.6 E
(against flows with small packets \(where the allowed sending)127.2
393.6 Q(rate is roughly a linear function of packet size\).)127.2 406.6
Q 6.6(4. A)100.8 432.6 R
(queue limited in bytes is better than a queue limited in)6.6 E
(packets for matching the worst-case queue size to the)127.2 445.6 Q
(worst-case queueing delay in seconds.)127.2 458.6 Q
(---------------------------------------------------------)100.8 471.6 Q
F2(Normati)72 503.6 Q .28 -.14(ve R)-.14 H(efer).14 E(ences)-.252 E F0
29.634([RFC2119] S.)113.1 520.2 R(Bradner)2.75 E 2.75(.K)-.605 G .33
-.165(ey W)-3.025 H(ords F)-.715 E
(or Use in RFCs to Indicate Requirement)-.165 E(Le)195.6 533.2 Q -.165
(ve)-.275 G(ls. RFC 2119.).165 E 29.634([RFC3448] M.)113.1 549.8 R
(Handle)2.75 E 1.43 -.715(y, S)-.165 H 2.75(.F).715 G(lo)-2.75 E
(yd, J. P)-.11 E(adh)-.165 E(ye, and J. W)-.055 E(idmer)-.44 E 2.75(,T)
-.44 G(CP Friendly Rate)-2.75 E
(Control \(TFRC\): Protocol Speci\214cation, RFC 3448, Proposed)195.6
562.8 Q(Standard, January 2003.)195.6 575.8 Q F2(Inf)72 601.8 Q(ormati)
-.35 E .28 -.14(ve R)-.14 H(efer).14 E(ences)-.252 E F0 46.761([EA03] W)
113.1 618.4 R 2.75(.E)-1.012 G(ddy and M. Allman.)-2.75 E 2.75(AC)5.5 G
(omparison of RED')-2.75 E 2.75(sB)-.605 G(yte and P)-2.75 E(ack)-.165 E
(et)-.11 E(Modes, Computer Netw)195.6 631.4 Q(orks, 42\(2\), June 2003.)
-.11 E 55.308([P04] T)113.1 648 R 2.75(.P)-.814 G
(helan, TFRC with Self-Limiting Sources, October 2004.)-2.75 E(URL)5.5 E
("http://www)195.6 661 Q(.phelan-4.com/dccp/".)-.715 E 35.134
([RFC791] J.)113.1 677.6 R
(Postel, Internet Protocol, RFC 791, September 1981.)2.75 E(Flo)72 733 Q
(yd/K)-.11 E 328.171(ohler [P)-.385 F(age 37])-.165 E 0 Cg EP
%%Page: 38 38
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E 35.134
([RFC879] J.)113.1 85 R(Postel, The TCP Maximum Se)2.75 E
(gment Size and Related T)-.165 E(opics, RFC)-.88 E(879, No)195.6 98 Q
-.165(ve)-.165 G(mber 1983.).165 E 29.634([RFC1144] V)113.1 114.6 R 2.75
(.J)-1.419 G(acobson, Compressing TCP/IP Headers for Lo)-2.75 E
(w-Speed Serial)-.275 E(Links, RFC 1144, February 1990.)195.6 127.6 Q
29.634([RFC3168] K.)113.1 144.2 R(Ramakrishnan, S. Flo)2.75 E
(yd, and D. Black, The Addition of Explicit)-.11 E
(Congestion Noti\214cation \(ECN\) to IP)195.6 157.2 Q 2.75(,R)-1.221 G
(FC 3168, September 2001.)-2.75 E 29.634([RFC3714] S.)113.1 173.8 R(Flo)
2.75 E(yd and J. K)-.11 E(empf, Editors.)-.275 E(IAB Concerns Re)5.5 E
-.055(ga)-.165 G(rding).055 E(Congestion Control for V)195.6 186.8 Q
(oice T)-1.419 E(raf)-.385 E(\214c in the Internet.)-.275 E(RFC 3714.)
5.5 E 29.634([RFC3819] P)113.1 203.4 R 2.75(.K)-1.221 G
(arn et al., Advice for Internet Subnetw)-2.75 E
(ork Designers, RFC 3819,)-.11 E(July 2004.)195.6 216.4 Q 29.634
([RFC4340] E.)113.1 233 R -.385(Ko)2.75 G(hler).385 E 2.75(,M)-.44 G
2.75(.H)-2.75 G(andle)-2.75 E 1.43 -.715(y, a)-.165 H(nd S. Flo).715 E
2.75(yd. Datagram)-.11 F(Congestion Control)2.75 E
(Protocol \(DCCP\), RFC 4340, March 2006.)195.6 246 Q 29.634
([RFC4342] S.)113.1 262.6 R(Flo)2.75 E(yd, E. K)-.11 E(ohler)-.385 E
2.75(,a)-.44 G(nd J. P)-2.75 E(adh)-.165 E 2.75(ye. Pro\214le)-.055 F
(for Datagram Congestion)2.75 E
(Control Protocol \(DCCP\) Congestion Control ID 3: TCP-Friendly)195.6
275.6 Q(Rate Control \(TFRC\).)195.6 288.6 Q(RFC 4342, March 2006.)5.5 E
16.797([RFC3448bis] M.)113.1 305.2 R(Handle)2.75 E 1.43 -.715(y, S)-.165
H 2.75(.F).715 G(lo)-2.75 E(yd, J. P)-.11 E(adh)-.165 E(ye, and J. W)
-.055 E(idmer)-.44 E 2.75(,T)-.44 G(CP Friendly Rate)-2.75 E(Control \(\
TFRC\): Protocol Speci\214cation, internet draft draft-ietf-dccp-)195.6
318.2 Q(rfc3448bis-00.txt, w)195.6 331.2 Q
(ork in progress, October 2006.)-.11 E 55.308([S05] Peter)113.1 347.8 R
(Sholander)2.75 E 2.75(,p)-.44 G(ri)-2.75 E -.275(va)-.275 G
(te communications, August 2005.).275 E(Citation for)5.5 E(ackno)195.6
360.8 Q(wledgement purposes only)-.275 E(.)-.715 E 53.482([V00] P)113.1
377.4 R 2.75(.V)-1.221 G 2.75(asallo. V)-3.971 F(ariable P)-1.221 E(ack)
-.165 E(et Size Equation-Based Congestion Control.)-.11 E(ICSI T)195.6
390.4 Q(echnical Report TR-00-008, April 2000.)-.77 E(URL)5.5 E
("http://www)195.6 403.4 Q(.icsi.berk)-.715 E(ele)-.11 E -.715(y.)-.165
G(edu/techreports/).715 E(2000.abstracts/tr)195.6 416.4 Q
(-00-008.html".)-.22 E([V)113.1 433 Q 21.527(OIPSIMS] W)-.44 F
(eb page "http://www)-.88 E(.icir)-.715 E(.or)-.605 E(g/tfrc/v)-.198 E
(oipsims.html".)-.22 E 36.982([WBL04] J.)113.1 449.6 R -.44(Wi)2.75 G
(dmer).44 E 2.75(,C)-.44 G 2.75(.B)-2.75 G(outremans, and Jean-Yv)-2.75
E(es Le Boudec.)-.165 E(Congestion)5.5 E(Control for Flo)195.6 462.6 Q
(ws with V)-.275 E(ariable P)-1.221 E(ack)-.165 E(et Size, A)-.11 E
(CM CCR, 34\(2\), 2004.)-.44 E/F1 14/Times-Bold@0 SF -.7(Au)72 488.6 S
(thors' Addr).7 E(esses)-.252 E F0(Sally Flo)100.8 505.2 Q(yd <\215o)
-.11 E(yd@icir)-.11 E(.or)-.605 E(g>)-.198 E
(ICSI Center for Internet Research)100.8 518.2 Q
(1947 Center Street, Suite 600)100.8 531.2 Q(Berk)100.8 544.2 Q(ele)-.11
E 1.43 -.715(y, C)-.165 H 2.75(A9).715 G(4704)-2.75 E(USA)100.8 557.2 Q
(Eddie K)100.8 573.8 Q(ohler <k)-.385 E(ohler@cs.ucla.edu>)-.11 E
(4531C Boelter Hall)100.8 586.8 Q(UCLA Computer Science Department)100.8
599.8 Q(Los Angeles, CA 90095)100.8 612.8 Q(USA)100.8 625.8 Q F1
(Full Copyright Statement)72 651.8 Q F0(Cop)100.8 668.4 Q
(yright \(C\) The Internet Society \(2006\).)-.11 E
(This document is subject to the rights, licenses)5.5 E
(and restrictions contained in BCP 78, and e)100.8 681.4 Q
(xcept as set forth therein, the authors retain all)-.165 E
(their rights.)100.8 694.4 Q(Flo)72 733 Q(yd/K)-.11 E 328.171(ohler [P)
-.385 F(age 38])-.165 E 0 Cg EP
%%Page: 39 39
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 78.61(-DRAFT Expires:)-1.012 F
(May 2007)2.75 E(No)102.986 E -.165(ve)-.165 G(mber 2006).165 E
(This document and the information contained herein are pro)100.8 85 Q
(vided on an "AS IS" basis and)-.165 E(THE CONTRIB)100.8 98 Q(UT)-.11 E
(OR, THE ORGANIZA)-.198 E(TION HE/SHE REPRESENTS OR IS)-1.221 E
(SPONSORED BY \(IF ANY\), THE INTERNET SOCIETY AND THE INTERNET)100.8
111 Q(ENGINEERING T)100.8 124 Q(ASK FORCE DISCLAIM ALL W)-1.023 E
(ARRANTIES, EXPRESS OR)-1.32 E(IMPLIED, INCLUDING B)100.8 137 Q(UT NO)
-.11 E 2.75(TL)-.44 G(IMITED T)-2.75 E 2.75(OA)-.198 G(NY W)-2.75 E
(ARRANTY THA)-1.32 E 2.75(TT)-1.221 G(HE USE)-2.75 E(OF THE INFORMA)
100.8 150 Q(TION HEREIN WILL NO)-1.221 E 2.75(TI)-.44 G
(NFRINGE ANY RIGHTS OR ANY)-2.75 E(IMPLIED W)100.8 163 Q
(ARRANTIES OF MERCHANT)-1.32 E(ABILITY OR FITNESS FOR A)-1.023 E -1.012
(PA)100.8 176 S -.66(RT)1.012 G(ICULAR PURPOSE.).66 E/F1 14/Times-Bold@0
SF(Intellectual Pr)72 202 Q(operty)-.252 E F0(The IETF tak)100.8 218.6 Q
(es no position re)-.11 E -.055(ga)-.165 G(rding the v).055 E
(alidity or scope of an)-.275 E 2.75(yI)-.165 G(ntellectual Property)
-2.75 E(Rights or other rights that might be claimed to pertain to the \
implementation or use of the)100.8 231.6 Q
(technology described in this document or the e)100.8 244.6 Q
(xtent to which an)-.165 E 2.75(yl)-.165 G(icense under such rights)
-2.75 E(might or might not be a)100.8 257.6 Q -.275(va)-.22 G
(ilable; nor does it represent that it has made an).275 E 2.75(yi)-.165
G(ndependent)-2.75 E(ef)100.8 270.6 Q(fort to identify an)-.275 E 2.75
(ys)-.165 G(uch rights.)-2.75 E
(Information on the procedures with respect to rights in)5.5 E
(RFC documents can be found in BCP 78 and BCP 79.)100.8 283.6 Q
(Copies of IPR disclosures made to the IETF Secretariat and an)100.8
300.2 Q 2.75(ya)-.165 G(ssurances of licenses to)-2.75 E(be made a)100.8
313.2 Q -.275(va)-.22 G(ilable, or the result of an attempt made to obt\
ain a general license or).275 E(permission for the use of such propriet\
ary rights by implementers or users of this)100.8 326.2 Q(speci\214cati\
on can be obtained from the IETF on-line IPR repository at)100.8 339.2 Q
(http://www)100.8 352.2 Q(.ietf.or)-.715 E(g/ipr)-.198 E(.)-.605 E
(The IETF in)100.8 368.8 Q(vites an)-.44 E 2.75(yi)-.165 G
(nterested party to bring to its attention an)-2.75 E 2.75(yc)-.165 G
(op)-2.75 E(yrights, patents or)-.11 E
(patent applications, or other proprietary rights that may co)100.8
381.8 Q -.165(ve)-.165 G 2.75(rt).165 G(echnology that may be)-2.75 E
(required to implement this standard.)100.8 394.8 Q
(Please address the information to the IETF at ietf-)5.5 E(ipr@ietf.or)
100.8 407.8 Q(g.)-.198 E(Flo)72 733 Q(yd/K)-.11 E 328.171(ohler [P)-.385
F(age 39])-.165 E 0 Cg EP
%%Trailer
end
%%EOF