One document matched: draft-ietf-dccp-tfrc-voip-04.ps

%!PS-Adobe-3.0
%%Creator: groff version 1.19
%%CreationDate: Tue Jan 24 16:16:45 2006
%%DocumentNeededResources: font Courier-Bold
%%+ font Times-Bold
%%+ font Times-Roman
%%+ font Courier
%%DocumentSuppliedResources: procset grops 1.19 0
%%Pages: 26
%%PageOrder: Ascend
%%DocumentMedia: Default 612 792 0 () ()
%%Orientation: Portrait
%%EndComments
%%BeginDefaults
%%PageMedia: Default
%%EndDefaults
%%BeginProlog
%%BeginResource: procset grops 1.19 0
/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{
/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
/setcmykcolor where{
pop
/Fk{
setcmykcolor fill
}bind def
}if
/Fg{
setgray fill
}bind def
/FL/fill load def
/LW/setlinewidth load def
/Cr/setrgbcolor load def
/setcmykcolor where{
pop
/Ck/setcmykcolor load def
}if
/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
/setpagedevice{}def
}bind def
/PEND{
clear
countdictstack CNT sub{end}repeat
level1 restore
}bind def
end def
/setpacking where{
pop
setpacking
}if
%%EndResource
%%BeginFeature: *PageSize Default
<< /PageSize [ 612 792 ] /ImagingBBox null >> setpagedevice
%%EndFeature
%%IncludeResource: font Courier-Bold
%%IncludeResource: font Times-Bold
%%IncludeResource: font Times-Roman
%%IncludeResource: font Courier
grops begin/DEFS 1 dict def DEFS begin/u{.001 mul}bind def end/RES 72
def/PL 792 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-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)179.999 E 317.999(INTERNET-DRAFT ICIR)72 98 R 167.999
(draft-ietf-dccp-tfrc-voip-04.ps Eddie)72 111 R(Kohler)6 E
(Expires: July 2006)72 124 Q(UCLA)299.999 E(24 January 2006)413.999 137
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)72 220
Q/F3 11/Times-Roman@0 SF
(By submitting this Internet-Draft, each author represents that an)100.8
236.6 Q 2.75(ya)-.165 G(pplicable patent or other)-2.75 E
(IPR claims of which he or she is a)100.8 249.6 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)100.8 262.6
Q -.11(wa)-.165 G
(re will be disclosed, in accordance with Section 6 of BCP 79.).11 E
(Internet-Drafts are w)100.8 279.2 Q
(orking documents of the Internet Engineering T)-.11 E(ask F)-.88 E
(orce \(IETF\), its)-.165 E(areas, and its w)100.8 292.2 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.)100.8 305.2 Q
(Internet-Drafts are draft documents v)100.8 321.8 Q
(alid for a maximum of six months and may be)-.275 E
(updated, replaced, or obsoleted by other documents at an)100.8 334.8 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)
100.8 347.8 Q(ork in progress.")-.11 E
(The list of current Internet-Drafts can be accessed at http://www)100.8
364.4 Q(.ietf.or)-.715 E(g/ietf/1id-)-.198 E(abstracts.txt.)100.8 377.4
Q(The list of Internet-Draft Shado)100.8 394 Q 2.75(wD)-.275 G
(irectories can be accessed at)-2.75 E(http://www)100.8 407 Q(.ietf.or)
-.715 E(g/shado)-.198 E -.715(w.)-.275 G(html.).715 E
(This Internet-Draft will e)100.8 423.6 Q(xpire on July 2006.)-.165 E F2
(Abstract)72 449.6 Q F3
(The proposal in this document is intended to be e)100.8 466.2 Q 2.75
(xperimental. More)-.165 F(speci\214cally)2.75 E 2.75(,t)-.715 G(his)
-2.75 E(proposal is intended for further e)100.8 479.2 Q
(xperimentation, b)-.165 E(ut not for widespread deplo)-.22 E
(yment at this)-.11 E(time in the global Internet.)100.8 492.2 Q(TCP-Fr\
iendly Rate Control \(TFRC\) is a congestion control mechanism for unic\
ast \215o)100.8 508.8 Q(ws)-.275 E(operating in a best-ef)100.8 521.8 Q
(fort Internet en)-.275 E(vironment [RFC 3448]. TFRC w)-.44 E
(as intended for)-.11 E(applications that use a \214x)100.8 534.8 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
547.8 Q(et size.)-.11 E(This)5.5 E(document proposes TFRC-SP)100.8 560.8
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)100.8 573.8 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 as a TCP \215o)100.8 586.8 Q 2.75(wu)
-.275 G(sing pack)-2.75 E(ets of up to 1500 bytes.)-.11 E
(TFRC-SP enforces a)5.5 E(Min Interv)100.8 599.8 Q
(al of 10 ms between data pack)-.275 E(ets, to pre)-.11 E -.165(ve)-.275
G(nt a single \215o).165 E 2.75(wf)-.275 G(rom sending small)-2.75 E
(pack)100.8 612.8 Q(ets arbitrarily frequently)-.11 E(.)-.715 E(Flo)
100.8 638.8 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)100.8 651.8 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.)100.8 664.8 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)100.8 677.8 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)100.8 690.8 Q .33 -.165(ve c)-.275 H
(onsiderably more than its share of the bandwidth.).165 E(Flo)72 769 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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 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)103.55 111 Q(oip-03.txt:)
-.22 E 2.75(*A)106.3 124 S
(dded a paragraph saying that this is intended for)-2.75 E
(Experimental, for further e)111.8 137 Q(xperimentation and not)-.165 E
(for widespread deplo)111.8 150 Q(yment.)-.11 E 2.75(*E)106.3 163 S
(diting of abstract so that it still \214ts the 25-line)-2.75 E(limit.)
111.8 176 Q(Changes from draft-ietf-dccp-tfrc-v)103.55 202 Q
(oip-02.txt:)-.22 E 2.75(*C)106.3 215 S(hanged name from "V)-2.75 E
(oIP v)-1.419 E(ariant of TFRC" to "TFRC-SP".)-.275 E 2.75(*A)106.3 228
S(dded Section 4.5 on "The Nominal P)-2.75 E(ack)-.165 E
(et Size", discussing)-.11 E(possible dif)114.55 241 Q(ferences in pack)
-.275 E(et drop rates between small)-.11 E(and lar)114.55 254 Q(ge pack)
-.198 E(ets.)-.11 E 2.75(*A)106.3 267 S(dded te)-2.75 E
(xt to Section 5 on "A Comparison with RFC 3714".)-.165 E 2.75(*A)106.3
280 S(dded te)-2.75 E
(xt to Section 6 on "TFRC-SP with Applications that)-.165 E
(Modify the P)117.3 293 Q(ack)-.165 E(et Size")-.11 E 2.75(*A)106.3 306
S(dded simulations with small-pack)-2.75 E(et TCP \215o)-.11 E(ws.)-.275
E 2.75(*A)106.3 319 S(dded a Security Considerations section.)-2.75 E
2.75(*M)106.3 332 S(inor editing.)-2.75 E
(Changes from draft-ietf-dccp-tfrc-v)103.55 345 Q(oip-01.txt:)-.22 E
2.75(*A)106.3 358 S
(dded modi\214ed algorithm for calculating the loss e)-2.75 E -.165(ve)
-.275 G(nt rate,).165 E(for short interv)117.3 371 Q
(als with multiple pack)-.275 E(et drops.)-.11 E 2.75(*M)106.3 384 S
-.165(ove)-2.75 G 2.75(dF).165 G(aster Restart to a separate document.)
-2.915 E 2.75(*A)106.3 397 S
(dded simulations with a con\214gured byte drop rate.)-2.75 E 2.75(*A)
106.3 410 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.)111.8 423 Q 2.75(*A)106.3 436 S(dded a discussion of unf)
-2.75 E(airness for Drop-T)-.11 E(ail with a queue)-.88 E(in bytes.)
111.8 449 Q(Changes from draft-ietf-dccp-tfrc-v)103.55 462 Q
(oip-00.txt:)-.22 E 2.75(*A)106.3 475 S(dded more simulations.)-2.75 E
2.75(*A)106.3 488 S(dded a Related W)-2.75 E(ork section.)-.88 E(Flo)72
769 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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E/F1 13/Times-Bold@0 SF -1.196(Ta)
253.526 85 S(ble of Contents)1.196 E/F2 10/Times-Roman@0 SF(1. Con)100.8
110 Q -.15(ve)-.4 G(ntions).15 E F0 11(.......................)10.44 G
F2(4)11.5 E(2. Introduction)100.8 122 Q F0 11(.......................)
11.01 G F2(4)11.5 E(3. TFRC-SP Congestion Control)100.8 134 Q F0 11
(..................)8.62 G F2(5)11.5 E(4. TFRC-SP Discussion)100.8 146 Q
F0 11(.....................)2.1 G F2(6)11.5 E
(4.1. The TCP Throughput Equation)115.8 158 Q F0 11(................)
9.48 G F2(6)11.5 E(4.2. Accounting for Header Size)115.8 170 Q F0 11
(.................)8.54 G F2(6)11.5 E(4.3. The TFRC-SP Min Interv)115.8
182 Q(al)-.25 E F0 11(.................)10.42 G F2(7)11.5 E
(4.4. Counting P)115.8 194 Q(ack)-.15 E(et Losses)-.1 E F0 11
(..................)10.84 G F2(8)11.5 E(4.5. The Nominal P)115.8 206 Q
(ack)-.15 E(et Size)-.1 E F0 11(..................)5.02 G F2(8)11.5 E
(5. A Comparison with RFC 3714)100.8 218 Q F0 11(..................)6.41
G F2(10)6.5 E(6. TFRC-SP with Applications that Modify the P)100.8 230 Q
(ack)-.15 E(et Size)-.1 E F0 11(..........)12.5 G F2(10)6.5 E
(7. Simulation Results)100.8 242 Q F0 11(.....................)12.1 G F2
(11)6.5 E(7.1. Simulations with Con\214gured P)115.8 254 Q(ack)-.15 E
(et Drop Rates)-.1 E F0 11(...........)10.7 G F2(11)6.5 E
(7.2. Simulations with Con\214gured Byte Drop Rates)115.8 266 Q F0 11
(............)4.47 G F2(14)6.5 E(7.3. P)115.8 278 Q(ack)-.15 E
(et Dropping Beha)-.1 E(vior at Routers with Drop-T)-.2 E(ail Queues)-.8
E F0 11(........).21 G F2(16)6.5 E(7.4. P)115.8 290 Q(ack)-.15 E
(et Dropping Beha)-.1 E(vior at Routers with A)-.2 E(QM)-.55 E F0 11
(...........)7.86 G F2(19)6.5 E(8. General Discussion)100.8 302 Q F0 11
(.....................)9.91 G F2(22)6.5 E(9. Security Considerations)
100.8 314 Q F0 11(....................)5.87 G F2(23)6.5 E(10. IAN)100.8
326 Q 2.5(AC)-.35 G(onsiderations)-2.5 E F0 11(....................)9.56
G F2(23)6.5 E(11. Thanks)100.8 338 Q F0 11(........................)
12.26 G F2(23)6.5 E(A. Appendix: Related W)100.8 350 Q(ork on Small-P)
-.8 E(ack)-.15 E(et V)-.1 E(ariants of TFRC)-1.11 E F0 11(.........)9.3
G F2(23)6.5 E(B. Appendix: A Discussion of P)100.8 362 Q(ack)-.15 E
(et Size and P)-.1 E(ack)-.15 E(et Dropping)-.1 E F0 11(.........)6.81 G
F2(24)6.5 E(Normati)100.8 374 Q .3 -.15(ve R)-.25 H(eferences).15 E F0
11(.....................)8.66 G F2(25)6.5 E(Informati)100.8 386 Q .3
-.15(ve R)-.25 H(eferences).15 E F0 11(.....................)4.22 G F2
(25)6.5 E(Authors' Addresses)100.8 398 Q F0 11(......................)
5.05 G F2(26)6.5 E(Full Cop)100.8 410 Q(yright Statement)-.1 E F0 11
(....................)10.12 G F2(26)6.5 E(Intellectual Property)100.8
422 Q F0 11(......................)2.27 G F2(26)6.5 E F0(Flo)72 769 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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 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 [RFC 2119].)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(,aS)-1.221 G
(mall-P)-2.75 E(ack)-.165 E(et v)-.11 E
(ariant of TCP-friendly rate control)-.275 E(\(TFRC\) [RFC 3448].)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 [RFC 3448], 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 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 lar).165 E(ge-)-.198 E(pack)100.8 290.8 Q
(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, limited by a)-.11
E(minimum interv)100.8 411.4 Q(al between pack)-.275 E(ets of 10 ms.)
-.11 E(\(RFC 3448 refers to this v)5.5 E(ariant of TFRC as)-.275 E
(TFRC-PS, for applications that might v)100.8 424.4 Q(ary their pack)
-.275 E(et size in response to congestion.\))-.11 E(It)5.5 E
(will better support applications that do not w)100.8 437.4 Q
(ant their sending rates in bytes per second to)-.11 E(suf)100.8 450.4 Q
(fer from their use of small pack)-.275 E 2.75(ets. This)-.11 F -.275
(va)2.75 G(riant is restricted to applications that send).275 E(pack)
100.8 463.4 Q(ets no more than once e)-.11 E -.165(ve)-.275 G
(ry 10 ms \(the Min Interv).165 E 2.75(al\). Gi)-.275 F -.165(ve)-.275 G
2.75(nt).165 G(his restriction, TFRC-)-2.75 E(SP ef)100.8 476.4 Q(fecti)
-.275 E -.165(ve)-.275 G(ly calculates the TFRC f).165 E
(air rate as if the bottleneck restriction w)-.11 E(as in bytes per)-.11
E 2.75(second. Applications)100.8 489.4 R(using TFRC-SP could ha)2.75 E
.33 -.165(ve a \214)-.22 H -.165(xe).165 G 2.75(dp).165 G(ack)-2.75 E
(et size, or could v)-.11 E(ary their)-.275 E(pack)100.8 502.4 Q
(et size in response to congestion.)-.11 E(TFRC-SP is moti)100.8 519 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)100.8 532 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)
100.8 545 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
558 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)100.8 571 Q
(wing:)-.275 E("While the ideal w)129.6 587.6 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)129.6 600.6 Q(appropriately when the limitation is in \
pps, such an ideal is hard to achie)129.6 613.6 Q -.165(ve)-.275 G 2.75
(.W).165 G(e)-3.63 E -.11(wo)129.6 626.6 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.)129.6 639.6 Q
(In addition, we note that the current TCP)5.5 E
(congestion control mechanisms are themselv)129.6 652.6 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)129.6 665.6 Q -.165(ve)-.275 G
(rse path in pps.).165 E(While the TCP)5.5 E(mechanisms do pro)129.6
678.6 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)129.6 691.6 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)129.6 704.6 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)129.6 717.6 Q
(ork limitation in Bps rather than in pps in)-.11 E(Flo)72 769 Q(yd/K)
-.11 E 284.171(ohler Section)-.385 F 2.75(2. [P)2.75 F(age 4])-.165 E 0
Cg EP
%%Page: 5 5
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E
(considering the minimum sending rate of telephon)129.6 85 Q 2.75(yt)
-.165 G(raf)-2.75 E(\214c.")-.275 E -.385(Tr)100.8 101.6 S(anslating th\
e discussion in [RFC 3714] to the congestion control mechanisms of TFRC\
,).385 E(it seems acceptable to standardize a v)100.8 114.6 Q
(ariant of TFRC that allo)-.275 E(ws lo)-.275 E(w-bandwidth \215o)-.275
E(ws)-.275 E(sending small pack)100.8 127.6 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
140.6 Q(This is accomplished by TFRC-)5.5 E(SP)100.8 153.6 Q 2.75(,as)
-1.221 G(mall modi\214cation to TFRC, as described belo)-2.75 E -.715
(w.)-.275 G(Maintaining incenti)100.8 170.2 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)100.8 183.2 Q
(dangers of encouraging a lar)100.8 196.2 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
209.2 Q 2.75(ets. This)-.11 F
(is discussed in more detail in Section 4.3. In addition, as)2.75 E(ag)
100.8 222.2 Q(ain discussed in Section 4.3, TFRC-SP includes the limita\
tion of the Min Interv)-.055 E(al)-.275 E(between pack)100.8 235.2 Q
(ets of 10 ms.)-.11 E -.165(Pa)100.8 251.8 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
264.8 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)100.8 277.8 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.)100.8 290.8 Q(Calculating the loss e)100.8 307.4
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)100.8 320.4 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)100.8 333.4 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)100.8 346.4 Q(ets)-.11 E
(per round-trip time in the f)100.8 359.4 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)100.8 372.4 Q -.165(ve)-.275 G
(nt per round-trip time, and are still able to detect).165 E
(and respond to increasingly hea)100.8 385.4 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)100.8 398.4 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
411.4 Q 2.75(ys)-.165 G(mall)-2.75 E(pack)100.8 424.4 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)100.8 437.4 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)100.8 450.4 Q
(et drop rate during periods of high congestion.)-.11 E
(This is discussed in more detail)5.5 E(in the section belo)100.8 463.4
Q -.715(w.)-.275 G/F1 11/Times-Bold@0 SF(3.)72 489.4 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)100.8 506 Q -.165(ve)-.275 G
2.75(ni).165 G 2.75(nS)-2.75 G(ection 3.1 of [RFC 3448], which gi)-2.75
E -.165(ve)-.275 G(s).165 E(the allo)100.8 519 Q
(wed sending rate X in bytes per second as a function of the loss e)
-.275 E -.165(ve)-.275 G(nt rate, pack).165 E(et)-.11 E
(size, and round-trip time.)100.8 532 Q
([RFC 3448] speci\214es that the pack)5.5 E(et size s used in the)-.11 E
(throughput equation should be the pack)100.8 545 Q
(et size used by the application, or the estimated)-.11 E(mean pack)
100.8 558 Q(et size if there are v)-.11 E(ariations in the pack)-.275 E
(et size depending on the data.)-.11 E(This gi)5.5 E -.165(ve)-.275 G(s)
.165 E(rough f)100.8 571 Q(airness with TCP \215o)-.11 E
(ws using the same pack)-.275 E(et size.)-.11 E
(TFRC-SP changes this beha)100.8 587.6 Q(vior in the follo)-.22 E
(wing w)-.275 E(ays.)-.11 E 12.65<8354>100.8 604.2 S(he nominal pack)
-12.65 E(et size: The nominal pack)-.11 E
(et size s is set to 1460 bytes.)-.11 E -.165(Fo)5.5 G(llo).165 E(wing)
-.275 E([RFC 3714], this pro)117.3 617.2 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)117.3 630.2 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)117.3 643.2 Q(et drop rate.)-.11 E 12.65
<8354>100.8 659.8 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
672.8 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)117.3 685.8 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)117.3 698.8 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)117.3 711.8 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)117.3 724.8 Q(wed transmit rate X as follo)
-.275 E(ws:)-.275 E(Flo)72 769 Q(yd/K)-.11 E 284.171(ohler Section)-.385
F 2.75(3. [P)2.75 F(age 5])-.165 E 0 Cg EP
%%Page: 6 6
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E 2.75(X:)117.3 85 S 2.75(=X*s)
-2.75 G(_true / \(s_true + H\),)-2.75 E(where s_true is the true a)117.3
101.6 Q -.165(ve)-.22 G(rage data pack).165 E
(et size for the connection in bytes, e)-.11 E(xcluding)-.165 E
(the transport and netw)117.3 114.6 Q(ork headers.)-.11 E
(The H parameter is set to the constant 40 bytes.)117.3 131.2 Q
(Thus, if the TFRC-SP application used)5.5 E(40-byte data se)117.3 144.2
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)117.3 157.2 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
170.2 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
(estimate of H, based on the actual netw)117.3 183.2 Q
(ork and transport headers to be used on the)-.11 E(connection')117.3
196.2 Q 2.75(sp)-.605 G(ack)-2.75 E 2.75(ets. F)-.11 F(or e)-.165 E
(xample, a DCCP connection [DCCP] o)-.165 E -.165(ve)-.165 G 2.75(rI)
.165 G(Pv4, where data)-2.75 E(pack)117.3 209.2 Q
(ets use the DCCP-Data pack)-.11 E
(et type, and there are no IP or DCCP options, could)-.11 E
(set H to 20 + 12 = 32 bytes.)117.3 222.2 Q 12.65<834d>100.8 238.8 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 \(those at)-.275 E(most tw)117.3 251.8 Q 2.75(or)-.11 G(ound-trip \
times\), the loss rate is computed by counting the actual number of)
-2.75 E(pack)117.3 264.8 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)117.3 277.8 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)117.3 290.8 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.)117.3 303.8 Q
(In standard TFRC, the TFRC recei)117.3 320.4 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)117.3 333.4 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)117.3 346.4 Q(al with N pack)-.275
E(ets and K losses, standard TFRC calculates the size of)-.11 E
(that loss interv)117.3 359.4 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)
117.3 372.4 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)117.3 385.4 Q
(ets including K losses, the size of the loss interv)-.11 E
(al is calculated as)-.275 E(N/K, contrib)117.3 398.4 Q
(uting to a loss e)-.22 E -.165(ve)-.275 G
(nt rate of K/N instead of 1/N.).165 E 12.65<8341>100.8 415 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
428 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)117.3 441 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)
117.3 454 Q -.715(w.)-.275 G/F1 11/Times-Bold@0 SF(4.)72 480 Q/F2 14
/Times-Bold@0 SF(TFRC-SP Discussion)5.5 E F1(4.1.)72 506 Q/F3 13
/Times-Bold@0 SF(The TCP Thr)5.5 E(oughput Equation)-.234 E F0
(TFRC-SP uses the TCP throughput equation gi)100.8 522.6 Q -.165(ve)
-.275 G 2.75(ni).165 G 2.75(n[)-2.75 G(RFC 3448].)-2.75 E(As sho)5.5 E
(wn in T)-.275 E(able 1 of)-.88 E([RFC 3714], for high pack)100.8 535.6
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)
100.8 548.6 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)100.8 561.6 Q(en)100.8 574.6 Q(vironments \(e.g., with pack)
-.44 E(et drop rates greater than 10%\), we note that it w)-.11 E
(ould be)-.11 E(useful to ha)100.8 587.6 Q .33 -.165(ve a t)-.22 H
(hroughput equation with a some).165 E
(what more moderate sending rate for)-.275 E(pack)100.8 600.6 Q
(et drop rates of 40% and abo)-.11 E -.165(ve)-.165 G(.).165 E F1(4.2.)
72 626.6 Q F3(Accounting f)5.5 E(or Header Size)-.325 E F0
([RFC 3714] mak)100.8 643.2 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)100.8 656.2 Q
(ycles or in pack)-.165 E(ets per second \(pps\).)-.11 E(Ho)100.8 669.2
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
682.2 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)100.8 695.2 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)100.8 708.2 Q(et header in the allo)-.11 E
(wed sending rate.)-.275 E(Flo)72 769 Q(yd/K)-.11 E 275.921
(ohler Section)-.385 F 2.75(4.2. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E(As an e)100.8 85 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 98 Q
(This is because the TCP-friendly rate is reduced by a)5.5 E -.11(fa)
100.8 111 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)100.8 124 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
137 Q(gments w)-.165 E(ould reduce the)-.11 E(allo)100.8 150 Q
(wed sending rate to 3.12 Kbps of application data.)-.275 E(\(In f)5.5 E
(act, the Min Interv)-.11 E(al w)-.275 E(ould)-.11 E(pre)100.8 163 Q
-.165(ve)-.275 G(nt senders from achie).165 E
(ving these rates, since applications using TFRC-SP cannot send)-.275 E
(more than 100 pack)100.8 176 Q(ets per second.\))-.11 E(Unless it has \
a more precise estimate of the header size, TFRC-SP assumes 40 bytes fo\
r the)100.8 192.6 Q(header size, although the header could be lar)100.8
205.6 Q(ger \(due to IP options, IPv6, IP tunnels, and)-.198 E(the lik)
100.8 218.6 Q(e\) or smaller \(due to header compression, DCCP instead \
of UDP\) on the wire.)-.11 E(Requiring the use of the e)100.8 231.6 Q
(xact header size in bytes w)-.165 E
(ould require signi\214cant additional)-.11 E(comple)100.8 244.6 Q(xity)
-.165 E 2.75(,a)-.715 G(nd w)-2.75 E(ould ha)-.11 E .33 -.165(ve l)-.22
H(ittle additional bene\214t.).165 E(TFRC-SP')5.5 E 2.75(sd)-.605 G(ef)
-2.75 E(ault assumption of a)-.11 E(40-byte header is suf)100.8 257.6 Q
(\214cient to get a rough estimate of the throughput, and to gi)-.275 E
.33 -.165(ve t)-.275 H(he).165 E(application some incenti)100.8 270.6 Q
.33 -.165(ve n)-.275 H(ot to use unnecessarily-man).165 E 2.75(ys)-.165
G(mall pack)-2.75 E 2.75(ets. Because)-.11 F(we are)2.75 E
(only aiming at rough f)100.8 283.6 Q(airness, and at a rough incenti)
-.11 E .33 -.165(ve f)-.275 H(or applications, the def).165 E
(ault use of a)-.11 E
(40-byte header in the calculations of the header bandwidth seems suf)
100.8 296.6 Q(\214cient.)-.275 E/F1 11/Times-Bold@0 SF(4.3.)72 322.6 Q
/F2 13/Times-Bold@0 SF(The TFRC-SP Min Inter)5.5 E -.13(va)-.13 G(l).13
E F0(The header size calculation pro)100.8 339.2 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
352.2 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)100.8 365.2 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)100.8 378.2 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
391.2 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)100.8 404.2 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)100.8 417.2 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)100.8 430.2 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)100.8 443.2 Q(vide suf)-.165 E
(\214cient protection for pps netw)-.275 E(ork limitations.)-.11 E
(TFRC-SP)100.8 459.8 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)100.8 472.8 Q 2.75(ys)-.165 G
(mall pack)-2.75 E(ets.)-.11 E(One justi\214cation for the Min Interv)
100.8 489.4 Q
(al is the practical one that the applications that currently)-.275 E
-.11(wa)100.8 502.4 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)100.8 515.4
Q(fect current traf)-.275 E 2.75(\214c. A)-.275 F
(second justi\214cation is that there is)2.75 E
(no pressing need for best-ef)100.8 528.4 Q(fort traf)-.275 E
(\214c in the current Internet to send small pack)-.275 E(ets more)-.11
E(frequently than once e)100.8 541.4 Q -.165(ve)-.275 G
(ry 10 ms \(rather than taking the 10 ms delay at the sender).165 E 2.75
(,a)-.44 G(nd)-2.75 E(mer)100.8 554.4 Q(ging the tw)-.198 E 2.75(os)-.11
G(mall pack)-2.75 E(ets into one lar)-.11 E(ger one\).)-.198 E
(This 10 ms delay for mer)5.5 E(ging small)-.198 E(pack)100.8 567.4 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
580.4 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)100.8 593.4 Q
(ving less than 10 ms between pack)-.22 E(ets is outweighed by the)-.11
E(bene\214t to the netw)100.8 606.4 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)100.8 623 Q
(al causes TFRC-SP not to support applications sending small pack)-.275
E(ets v)-.11 E(ery)-.165 E(frequently)100.8 636 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)100.8 649 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)100.8 662 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)
100.8 675 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)100.8 688 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)100.8 701 Q(wed sending rate.)-.275
E(This seems not only acceptable, b)5.5 E(ut)-.22 E
(desirable for the global Internet.)100.8 714 Q(Flo)72 769 Q(yd/K)-.11 E
275.921(ohler Section)-.385 F 2.75(4.3. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E(What is to pre)100.8 85 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 98 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)100.8 111 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)100.8 124 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)100.8 137 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)100.8 150 Q(times of congestion, if that seems desirable.)
100.8 163 Q(And \215o)5.5 E(ws that open multiple parallel)-.275 E
(connections are subject to the incon)100.8 176 Q -.165(ve)-.44 G
(niences of reordering and the lik).165 E(e.)-.11 E/F1 11/Times-Bold@0
SF(4.4.)72 202 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)100.8 218.6 Q(ets per round-trip)-.11 E
(time in the f)100.8 231.6 Q
(ace of high congestion, with a steady-state with multiple pack)-.11 E
(ets dropped per)-.11 E(round-trip time.)100.8 244.6 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)100.8 257.6 Q
(et per round-trip time.)-.11 E(In addition, for)5.5 E(TCP with T)100.8
270.6 Q(ahoe, Ne)-.88 E(wReno, or SA)-.275 E
(CK congestion control mechanisms, the response to)-.44 E
(congestion is lar)100.8 283.6 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)100.8 300.2 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)100.8 313.2 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)100.8 326.2 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)100.8 342.8 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)100.8 355.8 Q -.165(ve)-.275
G(nt rate [WBL04].).165 E(Instead of sending one)5.5 E(lar)100.8 368.8 Q
(ge pack)-.198 E
(et per round-trip time, TFRC-SP could be sending N small pack)-.11 E
(ets \(where N)-.11 E(small pack)100.8 381.8 Q(ets equal one lar)-.11 E
(ge 1500-byte pack)-.198 E 2.75(et\). The)-.11 F
(loss measurement used with TFRC-)2.75 E
(SP has to be able to detect a connection that is consistently recei)
100.8 394.8 Q(ving multiple pack)-.275 E(et losses)-.11 E
(or marks per round-trip time, to allo)100.8 407.8 Q 2.75(wT)-.275 G
(FRC-SP to respond appropriately)-2.75 E(.)-.715 E(In TFRC-SP)100.8
424.4 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)100.8 437.4 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)100.8 450.4 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)100.8 463.4 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)100.8 476.4 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.)100.8 489.4 Q
(Thus, if eight successi)5.5 E .33 -.165(ve l)-.275 H(oss interv).165 E
(als are short loss)-.275 E(interv)100.8 502.4 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.)100.8 515.4 Q
F1(4.5.)72 541.4 Q F2(The Nominal P)5.5 E(ack)-.13 E(et Size)-.13 E F0
(The guidelines in Section 3 abo)100.8 558 Q .33 -.165(ve s)-.165 H
(ay that the nominal pack).165 E(et size s is set to 1460 bytes,)-.11 E
(pro)100.8 571 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>100.8 584 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)100.8 597
Q(ws the assumption that a TCP \215o)-.275 E 2.75(ww)-.275 G
(ith 1460-byte pack)-2.75 E(ets will ha)-.11 E .33 -.165(ve a h)-.22 H
(igher).165 E(sending rate than a TCP \215o)100.8 610 Q 2.75(ww)-.275 G
(ith smaller pack)-2.75 E 2.75(ets. While)-.11 F
(this assumption holds in an)2.75 E(en)100.8 623 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)100.8 636 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)100.8 649 Q(ets.)-.11 E(The table belo)100.8 665.6
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)
100.8 678.6 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)100.8 691.6 Q
(et drop rates up to 10% or so; for)-.11 E(this en)100.8 704.6 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)100.8 717.6 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(Flo)72 769 Q(yd/K)
-.11 E 275.921(ohler Section)-.385 F 2.75(4.5. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E(size of s bytes, the pack)100.8
85 Q
(et drop rate is roughly s*p1, producing a sending rate in bytes per)
-.11 E -.66(RT)100.8 98 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)100.8 111 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.)100.8 124 Q(Each ro)100.8 140.6 Q 2.75(wo)-.275 G 2.75(fT)-2.75
G(able 1 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>
100.8 153.6 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)100.8 166.6 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)100.8 179.6
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)100.8 192.6 Q
(et size, up to a pack)-.11 E(et size of 1460 bytes, corresponding to)
-.11 E 2.75(ap)100.8 205.6 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)100.8 218.6 Q(gime of)-.165 E -.165(ex)100.8 231.6 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/F1 11/Courier@0 SF 13.2
(Segment Packet)193.2 250.6 R(TCP Rates \(Kbps\))39.6 E(Size \(B\))193.2
263.6 Q 13.2(DropRate SendRate)13.2 F(Goodput)26.4 E 6.6
(-------- --------)193.2 276.6 R 19.8(-------- -------)19.8 F 33
(14 0.005)219.6 289.6 R 39.6(6.37 6.34)46.2 F 33(128 0.016 30.78 30.30)
213 302.6 R 33(256 0.028 46.54 44.96)213 315.6 R 33
(512 0.053 62.43 58.37)213 328.6 R 33(1460 0.134 94.15 80.02)206.4 341.6
R 33(4000 0.324 35.20 21.44)206.4 354.6 R 33(8000 0.531 15.36)206.4
367.6 R(5.76)46.2 E(Table 1: TCP Median Send Rate vs. Packet Size I:)
173.4 393.6 Q(Byte Drop Rate 0.0001)259.2 406.6 Q F0 -.88(Ta)100.8 425.6
S(ble 2 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)100.8 438.6 Q
(ws with increasing pack)-.275 E(et size up to a pack)-.11 E
(et size of 128 bytes,)-.11 E(corresponding to a pack)100.8 451.6 Q
(et drop rate of 16%.)-.11 E(After than, the TCP sending rate decreases)
5.5 E(and then increases ag)100.8 464.6 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)
100.8 477.6 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)100.8 490.6 Q
(ut the TCP goodput rate remains essentially)-.22 E 2.75(zero. This)
100.8 503.6 R(mak)2.75 E(es sense, as almost all pack)-.11 E
(ets that are sent are dropped.)-.11 E F1 13.2(Segment Packet)193.2
522.6 R(TCP Rates \(Kbps\))39.6 E(Size \(B\))193.2 535.6 Q 13.2
(DropRate SendRate)13.2 F(Goodput)26.4 E 6.6(-------- --------)193.2
548.6 R 19.8(-------- -------)19.8 F 33(14 0.053)219.6 561.6 R 39.6
(1.68 1.56)46.2 F 33(128 0.159)213 574.6 R 39.6(7.66 6.13)46.2 F 33
(256 0.248)213 587.6 R 39.6(6.21 4.32)46.2 F 33(512 0.402)213 600.6 R
39.6(1.84 1.11)46.2 F 33(1460 0.712)206.4 613.6 R 39.6(1.87 0.47)46.2 F
33(4000 0.870)206.4 626.6 R 39.6(3.20 0.00)46.2 F 33(8000 0.890)206.4
639.6 R 39.6(5.76 0.00)46.2 F
(Table 2: TCP Median Send Rate vs. Packet Size II:)173.4 665.6 Q
(Byte Drop Rate 0.001)259.2 678.6 Q F0(The TCP beha)100.8 697.6 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)100.8 710.6 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)100.8 723.6 Q(ets, it mak)-.11 E
(es more sense to consider an ideal goal of a TFRC-SP \215o)-.11 E(w)
-.275 E(Flo)72 769 Q(yd/K)-.11 E 275.921(ohler Section)-.385 F 2.75
(4.5. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E(achie)100.8 85 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 98 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)100.8 111 Q(wed transmit rate;)
-.275 E(this means simply that in)5.5 E -.275(eva)100.8 124 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)100.8 137 Q
(et size \(though still at most 1500 bytes\) for that en)-.11 E
(vironment.)-.44 E/F1 11/Times-Bold@0 SF(5.)72 163 Q/F2 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)100.8 179.6 Q
(airness, potential congestion collapse, and poor user)-.11 E
(quality that could occur with the deplo)100.8 192.6 Q
(yment of non-congestion-controlled IP telephon)-.11 E(y)-.165 E -.165
(ove)100.8 205.6 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)100.8 218.6 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
231.6 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)100.8 244.6 Q
(et drop rate, and discontinue sending for a)-.11 E
(period if the steady-state pack)100.8 257.6 Q
(et drop rate signi\214cantly e)-.11 E(xceeds the allo)-.165 E
(wed threshold for)-.275 E(that minimum sending rate.)100.8 270.6 Q
(In determining the allo)100.8 287.2 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)100.8 300.2 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 1460-byte-pack)100.8 313.2 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)100.8 326.2 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)
100.8 339.2 Q(ould be required to)-.11 E
(terminate or suspend when the persistent pack)100.8 352.2 Q
(et drop rate signi\214cantly e)-.11 E(xceeded 25%.)-.165 E(One limitat\
ion of the lack of \214ne-grained control in the minimal mechanism desc\
ribed in)100.8 368.8 Q(RFC 3714 is that an IP telephon)100.8 381.8 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)100.8 394.8 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)100.8 407.8 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
420.8 Q(vy pack)-.22 E(et drop rates in the netw)-.11 E(ork.)-.11 E
(Because RFC 3714 assumes that the allo)100.8 437.4 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 w)100.8
450.4 Q(ould use *with the same pack)-.11 E(et drop rate*, the)-.11 E
(minimal mechanism in RFC 3714 w)100.8 463.4 Q(ould not pro)-.11 E
(vide f)-.165 E(airness between TCP and IP)-.11 E(telephon)100.8 476.4 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 than)-.11 E(lar)100.8 489.4 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 mak).165 E 2.75(et)-.11 G(he)-2.75 E
(optimistic assumption that a lar)100.8 502.4 Q(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 pack).165 E(et drop)-.11 E(rate as the IP telephon)100.8
515.4 Q 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 .55 -.275(ow w)-2.75 H(ould).165 E(recei)100.8 528.4 Q .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 \215o)-.11 E -.715(w.)-.275 G F1(6.)72 554.4 Q
F2(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)100.8 571 Q
(ould be with applications that maintain a \214x)-.11 E(ed sending)-.165
E(rate in pack)100.8 584 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')100.8 597 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.)100.8 610 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)100.8 623 Q
(et size that w)-.11 E(ould be needed for the sending)-.11 E
(rate in bytes per second not to e)100.8 636 Q(xceed the allo)-.165 E
(wed sending rate.)-.275 E(In en)5.5 E(vironments where the)-.44 E(pack)
100.8 649 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)100.8 662 Q(et drop rate e)-.11 E
(xperienced by the \215o)-.165 E -.715(w.)-.275 G(There are man)100.8
678.6 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)100.8 691.6 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)100.8 704.6 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)100.8 717.6 Q(et size, to a)
-.11 E -.22(vo)-.22 G(id an unnecessary reduction in the).22 E(Flo)72
769 Q(yd/K)-.11 E 278.671(ohler Section)-.385 F 2.75(6. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E(pack)100.8 85 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 addre\
ssed in documents that are more application-speci\214c.)100.8 98 Q
(Similarly)100.8 114.6 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.)100.8 127.6 Q
(In this case, the application might not reduce the pack)5.5 E
(et size until the current)-.11 E(pack)100.8 140.6 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
153.6 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)100.8 166.6 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.)
100.8 179.6 Q/F1 11/Times-Bold@0 SF(7.)72 205.6 Q/F2 14/Times-Bold@0 SF
(Simulation Results)5.5 E F0(This section e)100.8 222.2 Q(xplores the p\
erformance of TFRC-PS in simulation scenarios with con\214gured)-.165 E
(pack)100.8 235.2 Q(et or byte drop rates, and in scenarios with a rang\
e of queue management mechanisms)-.11 E(at the congested link.)100.8
248.2 Q(The simulations e)5.5 E(xplore en)-.165 E
(vironments where standard TFRC)-.44 E
(signi\214cantly limits the throughput of small-pack)100.8 261.2 Q
(et \215o)-.11 E(ws, and TFRC-SP gi)-.275 E -.165(ve)-.275 G 2.75(st)
.165 G(he desired)-2.75 E 2.75(throughput. The)100.8 274.2 R
(simulations also e)2.75 E(xplore en)-.165 E
(vironments where standard TFRC allo)-.44 E(ws)-.275 E(small-pack)100.8
287.2 Q(et \215o)-.11 E(ws to recei)-.275 E .33 -.165(ve g)-.275 H
(ood performance, while TFRC-SP is o).165 E -.165(ve)-.165 G
(rly aggressi).165 E -.165(ve)-.275 G(.).165 E
(The general lessons from the simulations are as follo)100.8 303.8 Q
(ws.)-.275 E 12.65<8349>100.8 320.4 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)117.3 333.4 Q -.165(ve)-.275 G 2.75(sr).165 G
(oughly the desired sending rate \(Sections 7.1, 7.3\).)-2.75 E 12.65
<8349>100.8 350 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)117.3 363 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(\(Section 7.2\).)117.3 376 Q 12.65<8349>100.8 392.6 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)117.3 405.6 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)117.3 418.6 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)
117.3 431.6 Q(ge-pack)-.198 E(et TCP \215o)-.11 E
(ws \(Sections 7.2, 7.3, 7.4\).)-.275 E 12.65<8353>100.8 448.2 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)117.3 461.2 Q(ail queues in bytes, and with A)-.88 E
(QM mechanisms in byte mode)-.605 E(\(Sections 7.3, 7.4\).)117.3 474.2 Q
(Those who are not interested in the details of the simulations could p\
roceed directly to)100.8 490.8 Q(Section 8 on General Discussion.)100.8
503.8 Q(TFRC-SP has been added to the NS simulator)100.8 520.4 Q 2.75
(,a)-.44 G(nd is illustrated in the v)-2.75 E(alidation test)-.275 E
("./test-all-friendly" in the directory tcl/tests.)100.8 533.4 Q
(The simulation scripts for the simulations in)5.5 E
(this document are a)100.8 546.4 Q -.275(va)-.22 G
(ilable at "http://www).275 E(.icir)-.715 E(.or)-.605 E(g/tfrc/v)-.198 E
2.75(oipsims.html". There)-.22 F(is also a)2.75 E
(pointer to the document "Graphs for draft-ietf-dccp-tfrc-v)100.8 559.4
Q(oip-03", which has graphs)-.22 E(sho)100.8 572.4 Q
(wing the information in tables in this document.)-.275 E F1(7.1.)72
598.4 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 from simulations comparing the throughput)
100.8 615 Q(of standard \(SA)100.8 628 Q(CK\) 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)100.8 641 Q 2.75
(ws. In)-.275 F(these simulations we con\214gure the router to)2.75 E
(randomly drop or mark pack)100.8 654 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)100.8 667 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)100.8 680 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)72 769 Q(yd/K)-.11 E 270.421(ohler Section)-.385 F 2.75
(7.1. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 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 3: 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 3 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 pack).165 E(ets, in order to)-.11 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 4.18 -.715(w. H)-.275 H
-.275(ow).715 G -2.365 -.275(ev e).275 H .88 -.44(r, w).275 H(e).44 E
(don')100.8 481 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)100.8 494 Q
(ets during high)-.11 E(le)100.8 507 Q -.165(ve)-.275 G
(ls of congestion.).165 E(Flo)72 769 Q(yd/K)-.11 E 270.421
(ohler Section)-.385 F 2.75(7.1. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 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 4: 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 4 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 4 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 4 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 4 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)72 769 Q(yd/K)
-.11 E 270.421(ohler Section)-.385 F 2.75(7.1. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 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 5: 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 5 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 4, 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(7.2.)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)72 769 Q(yd/K)-.11 E 270.421(ohler Section)-.385
F 2.75(7.2. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 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 6: Sending Rate vs. Byte Drop Rate)153.6 332 Q F0 -.88(Ta)100.8
354.6 S(ble 6 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 6 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 6 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)232.8 490.6 Q
(TFRC Pkt)33 E 26.4(Byte DropRate)173.4 503.6 R 39.6(DropRate TCP/TFRC)
26.4 F 6.6(DropRate \(1460B)160.2 516.6 R 6.6(seg\) \(14B)6.6 F 13.2
(seg\) Pkt)6.6 F(Drop Ratio)6.6 E 6.6(-------- ----------- ---------)
160.2 529.6 R(--------------)19.8 E 26.4(0.00001 0.015)166.8 542.6 R
46.2(0.0006 26.59)46.2 F 33(0.0001 0.13)166.8 555.6 R 46.2(0.0056 24.94)
52.8 F 39.6(0.001 0.77)166.8 568.6 R 52.8(0.054 14.26)52.8 F 39.6
(0.005 0.99)166.8 581.6 R 66(0.24 4.08)52.8 F 46.2(0.01 1.00 0.43)166.8
594.6 R(2.32)72.6 E 46.2(0.05 1.00 0.94)166.8 607.6 R(1.05)72.6 E
(Table 7: Packet Drop Rate Ratio vs. Byte Drop Rate)153.6 633.6 Q F0
-.88(Ta)100.8 656.2 S(ble 7 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)100.8 669.2 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
682.2 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)100.8 695.2 Q
(ets, and a pack)-.11 E(et drop rate of 0.054, or 5.4%,)-.11 E
(for the 56-byte TFRC pack)100.8 708.2 Q(ets.)-.11 E(Flo)72 769 Q(yd/K)
-.11 E 270.421(ohler Section)-.385 F 2.75(7.2. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E(The right column of T)100.8 85 Q
(able 7 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 6 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 6 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 pack)-.165 E
(ets, and the)-.11 E(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 6, 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 6, with high congestion and with byte drop rates of)-.88 E
(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 4 and 5 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 7.3 and 7.4 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(7.3.)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(The def)100.8 691.6 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)
100.8 704.6 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)100.8 717.6 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(Flo)72 769 Q
(yd/K)-.11 E 270.421(ohler Section)-.385 F 2.75(7.3. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E(pack)100.8 85 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 98 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
111 Q(ely to be dropped [WBL04]? And what is the relationship between t\
he pack)-.11 E(et-dropping)-.11 E(beha)100.8 124 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)213 143
S(end Rates in Kbps - - - - >)-6.6 E 46.2(Web TCP)153.6 156 R
(\(1460B seg\))6.6 E(TFRC-SP \(200B seg\))33 E 13.2(Sessions DropRate)
140.4 169 R 19.8(SendRate DropRate)13.2 F(SendRate)13.2 E 13.2
(-------- --------)140.4 182 R 19.8(-------- --------)13.2 F(--------)
13.2 E 39.6(10 0.04)166.8 195 R 39.6(316.18 0.05)33 F(183.05)33 E 39.6
(25 0.07)166.8 208 R 39.6(227.47 0.07)33 F(181.23)33 E 39.6(50 0.08)
166.8 221 R 39.6(181.10 0.08)33 F(178.32)33 E 39.6(100 0.14)160.2 234 R
39.6(85.97 0.12)39.6 F(151.42)33 E 39.6(200 0.17)160.2 247 R 39.6
(61.20 0.14)39.6 F(73.88)39.6 E 39.6(400 0.20)160.2 260 R 39.6
(27.79 0.18)39.6 F(36.81)39.6 E 39.6(800 0.29 3.50 0.27)160.2 273 R
(16.33)39.6 E 39.6(1600 0.37 0.63 0.33 6.29)153.6 286 R
(Table 8: Drop and Send Rates for Drop-Tail Queues in Packets)127.2 312
Q F0 -.88(Ta)100.8 334.6 S(ble 8 sho).88 E(ws the results of the second\
 half of 100-second simulations, with \214v)-.275 E 2.75(eT)-.165 G(CP)
-2.75 E(connections and \214v)100.8 347.6 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)100.8 360.6 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.)
100.8 373.6 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
386.6 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)100.8 399.6 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)100.8 412.6 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)100.8 425.6
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)
100.8 438.6 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)100.8 451.6 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)100.8 464.6 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)100.8 477.6 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)100.8 490.6 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
503.6 Q(et size.)-.11 E -.88(Ta)100.8 520.2 S(ble 8 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 2.75<658d>-.165 G -.275(ow)-2.75 G(s.).275 E(As e)100.8 533.2 Q
(xpected, the TFRC-SP \215o)-.165 E(ws see similar pack)-.275 E
(et drop rates as the TCP \215o)-.11 E(ws, though the)-.275 E
(TFRC-SP \215o)100.8 546.2 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 rates of)-.11 E(25% or higher)100.8 559.2 Q(.)-.605 E(Flo)72
769 Q(yd/K)-.11 E 270.421(ohler Section)-.385 F 2.75(7.3. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 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(239.81 0.00)33 F(185.19)33 E 39.6
(25 0.09)180 150 R 39.6(189.02 0.01)33 F(184.95)33 E 39.6(50 0.14)180
163 R 39.6(99.46 0.01)39.6 F(185.07)33 E 39.6(100 0.20)173.4 176 R 39.6
(16.42 0.02)39.6 F(183.77)33 E 39.6(200 0.26 4.46 0.03)173.4 189 R
(181.98)33 E 39.6(400 0.29 4.61 0.05)173.4 202 R(151.88)33 E 39.6
(800 0.49 1.01 0.08)173.4 215 R(113.10)33 E 39.6(1600 0.65 0.67 0.12)
166.8 228 R(65.17)39.6 E
(Table 9: Drop and Send Rates for Drop-Tail Queues in Bytes I:)107.4 254
Q(1460B TCP Segments)272.4 267 Q F0(Ho)100.8 289.6 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)100.8 302.6
Q 2.75(ets. F)-.11 F(or a queue in pack)-.165 E(ets, the)-.11 E
(queue has a \214x)100.8 315.6 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)100.8 328.6 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)100.8 341.6 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,)100.8 354.6 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.)100.8 367.6 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)100.8 380.6 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
393.6 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)100.8 406.6 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(As T)100.8 423.2 Q
(able 9 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 smaller)-2.75 E(pack)100.8 436.2 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 sending)-.198 E 2.75(rate. F)100.8 449.2 R
(or the simulations in T)-.165 E(able 9, the TFRC-SP \215o)-.88 E
(ws use 200-byte data se)-.275 E(gments, while)-.165 E(the long-li)100.8
462.2 Q -.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, when the \214v)-.165 E 2.75(eT)-.165 G(CP)-2.75 E<8d6f>100.8
475.2 Q(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 \214v)-.275 E 2.75(eT)
-.165 G(CP \215o)-2.75 E(ws see)-.275 E(an a)100.8 488.2 Q -.165(ve)-.22
G(rage drop rate of 49% in the second half of the simulation, while the\
 \214v).165 E 2.75(eT)-.165 G(FRC-SP)-2.75 E<8d6f>100.8 501.2 Q
(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 .33 -.165(ve m)
-.275 H(ore than 100).165 E(times the throughput of the TCP \215o)100.8
514.2 Q 2.75(ws. This)-.275 F(raises serious questions about making the)
2.75 E(assumption that \215o)100.8 527.2 Q(ws with small pack)-.275 E
(ets see the same pack)-.11 E(et drop rate as \215o)-.11 E(ws with lar)
-.275 E(ger)-.198 E(pack)100.8 540.2 Q 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 realistic)-.055 E
(Internet scenarios, in terms of whether lar)100.8 553.2 Q(ge pack)-.198
E(ets are considerably more lik)-.11 E(ely to be)-.11 E
(dropped than are small ones.)100.8 566.2 Q(Flo)72 769 Q(yd/K)-.11 E
270.421(ohler Section)-.385 F 2.75(7.3. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 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 10: 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 10 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/F2 11
/Times-Bold@0 SF(7.4.)72 341.6 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)100.8 358.2 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 \(A)100.8 371.2 Q(QM\) 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)100.8 384.2 Q -.165(ve)-.275 G
(ral parameters than can af).165 E(fect the pack)-.275 E
(et dropping or marking)-.11 E(beha)100.8 397.2 Q
(vior as a function of the pack)-.22 E(et size.)-.11 E
(First, as with Drop-T)100.8 413.8 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)100.8 426.8 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)100.8 439.8 Q -.165(ve)-.165 G
(r\215o).165 E(ws.)-.275 E(Second, and orthogonally)100.8 456.4 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)100.8 469.4 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
482.4 Q(In pack)5.5 E(et)-.11 E(mode, lar)100.8 495.4 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)100.8 508.4 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)100.8 521.4 Q .33
-.165(ve r)-.275 H(oughly the same *number* of pack).165 E(et drops.)
-.11 E(The simulations reported in the)5.5 E(appendix sho)100.8 534.4 Q
2.75(wt)-.275 G(hat for RED in pack)-2.75 E(et mode, the pack)-.11 E
(et drop rates for the TFRC-SP \215o)-.11 E(ws)-.275 E
(are similar to those for the TCP \215o)100.8 547.4 Q
(ws, with a resulting acceptable throughput for the)-.275 E
(TFRC-SP \215o)100.8 560.4 Q 5.5(ws. This)-.275 F
(is true with the queue in pack)2.75 E
(ets or in bytes, and with or without)-.11 E(Adapti)100.8 573.4 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 between TCP and)-.11 E(TFRC-SP \215o)100.8 586.4 Q
(ws does not hold for RED in byte mode.)-.275 E
(The third RED parameter that af)100.8 603 Q(fects the pack)-.275 E
(et dropping or marking beha)-.11 E(vior as a)-.22 E(function of pack)
100.8 616 Q
(et size is whether the RED mechanism is using Standard RED or Adapti)
-.11 E -.165(ve)-.275 G 2.75(RED; Adapti)100.8 629 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)
100.8 642 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)
100.8 655 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)100.8 668 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)100.8 681 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
694 Q 2.75(gimes. One)-.165 F
(consequence is that when byte mode is used with Adapti)100.8 707 Q .33
-.165(ve R)-.275 H(ED, the byte mode e).165 E(xtends)-.165 E -2.365
-.275(ev e)100.8 720 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(Flo)72 769 Q(yd/K)-.11 E 270.421
(ohler Section)-.385 F 2.75(7.4. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E
(byte mode is no longer in use when the drop rate e)100.8 85 Q
(xceeds the \214x)-.165 E(ed dropping threshold \(set)-.165 E(by def)
100.8 98 Q(ault to 10% in the NS simulator\).)-.11 E
(In the simulations in this section, we e)100.8 114.6 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.)100.8 127.6 Q
(In this simulations, as in Section 7.3 abo)5.5 E -.165(ve)-.165 G 2.75
(,t).165 G(he application for the)-2.75 E(TFRC-SP \215o)100.8 140.6 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)226.2 172.6
S(end Rates in Kbps - - - - >)-6.6 E 46.2(Web TCP)166.8 185.6 R
(\(1460B seg\))6.6 E(TFRC-SP \(200B seg\))33 E 13.2(Sessions DropRate)
153.6 198.6 R 19.8(SendRate DropRate)13.2 F(SendRate)13.2 E 13.2
(-------- --------)153.6 211.6 R 19.8(-------- --------)13.2 F(--------)
13.2 E 39.6(10 0.05)180 224.6 R 39.6(305.76 0.04)33 F(182.82)33 E 39.6
(25 0.06)180 237.6 R 39.6(224.16 0.06)33 F(175.91)33 E 39.6(50 0.09)180
250.6 R 39.6(159.12 0.08)33 F(152.51)33 E 39.6(100 0.13)173.4 263.6 R
39.6(90.77 0.11)39.6 F(106.13)33 E 39.6(200 0.14)173.4 276.6 R 39.6
(48.53 0.14)39.6 F(70.25)39.6 E 39.6(400 0.20)173.4 289.6 R 39.6
(22.08 0.19)39.6 F(41.50)39.6 E 39.6(800 0.27 3.55 0.25)173.4 302.6 R
(17.50)39.6 E 39.6(1600 0.42 1.87 0.34 8.81)166.8 315.6 R
(Table 11: Drop and Send Rates for RED Queues in Packet Mode)114 341.6 Q
F0 -.165(Fo)100.8 364.2 S 2.75(rt).165 G(he simulations in T)-2.75 E
(able 11, 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
377.2 Q(ail queue in pack)-.88 E(ets, as in T)-.11 E(able 8 abo)-.88 E
-.165(ve)-.165 G 5.5(.T).165 G(he)-5.5 E(TFRC-SP \215o)100.8 390.2 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 2.75(sh).165 G
(igher)-2.75 E(throughput in the more congested en)100.8 403.2 Q 2.75
(vironments. The)-.44 F(simulations are similar with a RED)2.75 E
(queue in pack)100.8 416.2 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)2.75 E
(simulations, TFRC-SP gi)100.8 429.2 Q -.165(ve)-.275 G 2.75(sr).165 G
(oughly the desired performance.)-2.75 E F1 6.6(<-----S)226.2 448.2 S
(end Rates in Kbps - - - - >)-6.6 E 39.6(Web TCP)166.8 461.2 R
(\(1460B seg\))6.6 E(TFRC-SP \(200B seg\))39.6 E 13.2(Sessions DropRate)
153.6 474.2 R 19.8(SendRate DropRate)13.2 F(SendRate)13.2 E 13.2
(-------- --------)153.6 487.2 R 19.8(-------- --------)13.2 F(--------)
13.2 E 39.6(10 0.06)180 500.2 R 39.6(272.16 0.02)33 F(184.37)33 E 39.6
(25 0.07)180 513.2 R 39.6(175.82 0.02)33 F(184.06)33 E 39.6(50 0.10)180
526.2 R 39.6(75.65 0.04)39.6 F(180.56)33 E 39.6(100 0.14)173.4 539.2 R
39.6(38.98 0.07)39.6 F(151.65)33 E 39.6(200 0.19)173.4 552.2 R 39.6
(16.66 0.11)39.6 F(106.80)33 E 39.6(400 0.26 4.85 0.15)173.4 565.2 R
(69.41)39.6 E 39.6(800 0.35 3.12 0.20)173.4 578.2 R(27.07)39.6 E 39.6
(1600 0.42 0.67 0.29)166.8 591.2 R(10.68)39.6 E
(Table 12: Drop and Send Rates for RED Queues in Byte Mode)127.2 617.2 Q
F0 -.88(Ta)100.8 639.8 S(ble 12 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)100.8 652.8 Q(what greater dif)-.275 E
(ferent between the pack)-.275 E
(et drop rates between the TCP and TFRC-)-.11 E(SP \215o)100.8 665.8 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 12, the)
-.88 E(pack)100.8 678.8 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)
100.8 691.8 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)100.8 704.8
Q(fected in the lo)-.275 E(wer pack)-.275 E(et-drop-)-.11 E(rate re)
100.8 717.8 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(Flo)72 769 Q(yd/K)-.11 E 270.421(ohler Section)-.385 F 2.75
(7.4. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E(or more.)100.8 85 Q
(The sending rate for TFRC-SP in the scenarios in T)5.5 E
(able 12 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
98 Q(.)-.605 E(Ho)100.8 111 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
124 Q/F1 11/Courier@0 SF 6.6(<-----S)226.2 156 S
(end Rates in Kbps - - - - >)-6.6 E 46.2(Web TCP)166.8 169 R
(\(512B seg\))6.6 E(TFRC-SP \(200B seg\))39.6 E 13.2(Sessions DropRate)
153.6 182 R 19.8(SendRate DropRate)13.2 F(SendRate)13.2 E 13.2
(-------- --------)153.6 195 R 19.8(-------- --------)13.2 F(--------)
13.2 E 39.6(10 0.01)180 208 R 39.6(337.86 0.01)33 F(184.06)33 E 39.6
(25 0.02)180 221 R 39.6(258.71 0.01)33 F(184.03)33 E 39.6(50 0.02)180
234 R 39.6(184.71 0.01)33 F(183.99)33 E 39.6(100 0.04)173.4 247 R 39.6
(63.63 0.03)39.6 F(184.43)33 E 39.6(200 0.08)173.4 260 R 39.6
(28.95 0.06)39.6 F(149.80)33 E 39.6(400 0.12)173.4 273 R 39.6
(17.03 0.10)39.6 F(88.21)39.6 E 39.6(800 0.24 5.94 0.15)173.4 286 R
(36.80)39.6 E 39.6(1600 0.32 3.37 0.21)166.8 299 R(19.45)39.6 E
(Table 13: Drop and Send Rates for RED Queues in Byte Mode)127.2 325 Q
F0 -.88(Ta)100.8 347.6 S(ble 13 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>100.8 360.6 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)100.8 373.6 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)100.8 386.6 Q
(able 13 with higher pack)-.88 E(et drop rates are)-.11 E
(still greater than desired, e)100.8 399.6 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
412.6 Q(gments instead of 512-byte data se)-.165 E(gments.)-.165 E F1
6.6(<-----S)226.2 444.6 S(end Rates in Kbps - - - - >)-6.6 E 39.6
(Web TCP)166.8 457.6 R(\(1460B seg\))6.6 E(TFRC-SP \(200B seg\))39.6 E
13.2(Sessions DropRate)153.6 470.6 R 19.8(SendRate DropRate)13.2 F
(SendRate)13.2 E 13.2(-------- --------)153.6 483.6 R 19.8
(-------- --------)13.2 F(--------)13.2 E 39.6(10 0.04)180 496.6 R 39.6
(318.10 0.02)33 F(185.34)33 E 39.6(25 0.08)180 509.6 R 39.6(175.34 0.03)
33 F(184.38)33 E 39.6(50 0.10)180 522.6 R 39.6(81.60 0.04)39.6 F(181.95)
33 E 39.6(100 0.12)173.4 535.6 R 39.6(28.51 0.05)39.6 F(178.79)33 E 39.6
(200 0.20 3.65 0.06)173.4 548.6 R(173.78)33 E 39.6(400 0.27 1.44 0.08)
173.4 561.6 R(161.41)33 E 39.6(800 0.40 0.58 0.06)173.4 574.6 R(159.62)
33 E 39.6(1600 0.55 0.29 0.02)166.8 587.6 R(180.92)33 E
(Table 14: Drop and Send Rates with Adaptive RED Queues in Byte Mode)
100.8 613.6 Q F0 -.165(Fo)100.8 636.2 S 2.75(rt).165 G
(he simulations in T)-2.75 E
(able 14, the congested router uses an Adapti)-.88 E .33 -.165(ve R)
-.275 H(ED queue in byte).165 E(mode.)100.8 649.2 Q -.165(Fo)100.8 665.8
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
678.8 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)100.8 691.8 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)100.8 708.4 Q(able 14 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)100.8 721.4 Q(\214c is *much* lo)-.275 E
(wer than that for the TCP traf)-.275 E(\214c, and as a)-.275 E(Flo)72
769 Q(yd/K)-.11 E 270.421(ohler Section)-.385 F 2.75(7.4. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E
(consequence, the sending rate for the TFRC-SP traf)100.8 85 Q
(\214c in a highly congested en)-.275 E(vironment is)-.44 E
(*much* higher than that of the TCP traf)100.8 98 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)100.8 111 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 1.76 -.88(We n)100.8 127.6 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(ng).165 G(reater than)-2.75 E(the unf)100.8 140.6 Q
(airness sho)-.11 E(wn in T)-.275 E(able 9 for a Drop-T)-.88 E
(ail queue in bytes.)-.88 E(At the same time, it is not)5.5 E(kno)100.8
153.6 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 byte).165 E(mode, or of an)100.8
166.6 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 of the)-.165 E(deplo)100.8 179.6 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/F1 11/Courier@0 SF 6.6
(<-----S)226.2 198.6 S(end Rates in Kbps - - - - >)-6.6 E 46.2(Web TCP)
166.8 211.6 R(\(512B seg\))6.6 E(TFRC-SP \(200B seg\))39.6 E 13.2
(Sessions DropRate)153.6 224.6 R 19.8(SendRate DropRate)13.2 F(SendRate)
13.2 E 13.2(-------- --------)153.6 237.6 R 19.8(-------- --------)13.2
F(--------)13.2 E 39.6(10 0.01)180 250.6 R 39.6(306.56 0.01)33 F(185.11)
33 E 39.6(25 0.02)180 263.6 R 39.6(261.41 0.01)33 F(184.41)33 E 39.6
(50 0.02)180 276.6 R 39.6(185.07 0.01)33 F(184.54)33 E 39.6(100 0.04)
173.4 289.6 R 39.6(59.25 0.03)39.6 F(181.58)33 E 39.6(200 0.08)173.4
302.6 R 39.6(16.32 0.06)39.6 F(150.87)33 E 39.6(400 0.12)173.4 315.6 R
39.6(11.53 0.10)39.6 F(98.10)39.6 E 39.6(800 0.25 5.85 0.15)173.4 328.6
R(46.59)39.6 E 39.6(1600 0.32 1.43 0.22)166.8 341.6 R(19.40)39.6 E
(Table 15: Drop and Send Rates for Adaptive RED Queues in Byte Mode)
100.8 367.6 Q F0 -.88(Ta)100.8 390.2 S(ble 15 sho).88 E
(ws that when TFRC-SP with 240-byte pack)-.275 E
(ets competes with TCP with)-.11 E(552-byte pack)100.8 403.2 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
416.2 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)100.8 429.2 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)100.8 442.2 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)100.8 455.2 Q
(fecti)-.275 E .33 -.165(ve t)-.275 H(hat than in T).165 E(able 14 abo)
-.88 E -.165(ve)-.165 G(.).165 E/F2 11/Times-Bold@0 SF(8.)72 481.2 Q/F3
14/Times-Bold@0 SF(General Discussion)5.5 E F0
(Dropping rates for small pack)100.8 497.8 Q
(ets: The goal of TFRC-SP is for small-pack)-.11 E(et TFRC-SP \215o)-.11
E(ws)-.275 E(to ha)100.8 510.8 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)
100.8 523.8 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)100.8 536.8 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)100.8 549.8 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)100.8 562.8 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)100.8 575.8 Q
(ge-pack)-.198 E(et TCP \215o)-.11 E(ws.)-.275 E(The accurac)100.8 592.4
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 \(i.e., pack)100.8 605.4 Q
(ets of at most 120 bytes\) TFRC-SP)-.11 E
(only restricts the sending rate when the pack)100.8 618.4 Q
(et drop rate is f)-.11 E(airly high, e.g., greater than)-.11 E 2.75
(10%. [Deri)100.8 631.4 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)100.8 644.4 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)100.8 657.4 Q -.275(va)-.275 G(lent to a TCP \215o).275 E
2.75(ww)-.275 G(ith 1460-byte pack)-2.75 E(ets sending 2.2 pack)-.11 E
(ets per round-trip)-.11 E 2.75(time. From)100.8 670.4 R -.88(Ta)2.75 G
(ble 1 of RFC 3714, this sending rate can be sustained with a pack).88 E
(et drop rate)-.11 E(slightly greater than 10%.])100.8 683.4 Q
(In this high-pack)5.5 E(et-drop re)-.11 E
(gime, the performance of TFRC-SP)-.165 E
(is determined in part by the accurac)100.8 696.4 Q 2.75(yo)-.165 G 2.75
(ft)-2.75 G(he TCP response function in representing the)-2.75 E
(actual sending rate of a TCP connection.)100.8 709.4 Q(Flo)72 769 Q
(yd/K)-.11 E 278.671(ohler Section)-.385 F 2.75(8. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E(In this 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 timestamps and/or of ECN, and the lik).385 E 2.75
(e. It)-.11 F(is good to insure that)2.75 E(simulations or e)100.8 124 Q
(xperiments e)-.165 E(xploring f)-.165 E(airness include the e)-.11 E
(xploration of f)-.165 E(airness with the)-.11 E(most aggressi)100.8 137
Q .33 -.165(ve T)-.275 H
(CP mechanisms conformance with the current standards.).165 E(Our)5.5 E
(simulations use SA)100.8 150 Q(CK TCP with Limited T)-.44 E
(ransmit and with a minimum R)-.385 E .396 -.198(TO o)-.66 H 2.75(f2)
.198 G(00 ms.)-2.75 E
(Adding the use of timestamps has not made a big dif)100.8 163 Q 2.75
(ference. W)-.275 F 2.75(eh)-.88 G -2.475 -.22(av e)-2.75 H(n').22 E
2.75(tu)-.198 G(sed TCP with)-2.75 E
(ECN, because our judgment is that in high pack)100.8 176 Q(et drop re)
-.11 E(gimes, it is preferable for A)-.165 E(QM)-.605 E
(mechanisms to drop rather than mark pack)100.8 189 Q(ets.)-.11 E(Gener\
al issues for TFRC: The congestion control mechanisms in TFRC and TFRC-\
SP)100.8 205.6 Q(limit a \215o)100.8 218.6 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)100.8 231.6 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)100.8 244.6 Q
(ursty TCP web traf)-.22 E(\214c, in scenarios with lo)-.275 E(w)-.275 E
(le)100.8 257.6 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.)72 283.6 Q
/F2 14/Times-Bold@0 SF(Security Considerations)5.5 E F0
(There are no security considerations introduced in this document.)100.8
300.2 Q
(General security considerations for TFRC are discussed in RFC 3448.)
100.8 316.8 Q(The security)5.5 E
(considerations for TFRC include the need to protect ag)100.8 329.8 Q
(ainst spoofed feedback, and the)-.055 E(need for protection mechanisms\
 to protect the congestion control mechanisms ag)100.8 342.8 Q(ainst)
-.055 E(incorrect information from the recei)100.8 355.8 Q -.165(ve)
-.275 G -.605(r.).165 G(Security considerations for DCCP')100.8 372.4 Q
2.75(sC)-.605 G(ongestion Control ID 3, TFRC Congestion Control,)-2.75 E
(are discussed in [CCID 3 PR)100.8 385.4 Q 2.75(OFILE]. That)-.44 F
(document e)2.75 E(xtensi)-.165 E -.165(ve)-.275 G(ly discussed the).165
E(mechanisms the sender can use to v)100.8 398.4 Q
(erify the information sent by the recei)-.165 E -.165(ve)-.275 G -.605
(r.).165 G F1(10.)72 424.4 Q F2(IAN)5.5 E 3.5(AC)-.28 G(onsiderations)
-3.5 E F0(There are no IAN)100.8 441 Q 2.75(Ac)-.385 G
(onsiderations in this document.)-2.75 E F1(11.)72 467 Q F2(Thanks)5.5 E
F0 1.76 -.88(We t)100.8 483.6 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)
100.8 496.6 Q 2.75(ws. W)-.275 F 2.75(et)-.88 G(hank Joer)-2.75 E 2.75
(gW)-.198 G(idmer for feedback on earlier)-3.19 E -.165(ve)100.8 509.6 S
(rsions of this draft.).165 E 1.76 -.88(We a)5.5 H(lso thank the DCCP W)
.88 E(orking Group for feedback and)-.88 E(discussions.)100.8 522.6 Q F1
(A.)72 548.6 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)100.8 565.2 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)100.8 578.2 Q(ws are not)-.275 E
(penalized by sending smaller pack)100.8 591.2 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
604.2 Q(et size used by TFRC, and counting)-.11 E(the pack)100.8 617.2 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)100.8 630.2 Q
-.165(ve)-.275 G 2.75(dp).165 G(ack)-2.75 E(ets into MTU-sized units.)
-.11 E([WBL04] also ar)100.8 646.8 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)100.8 659.8 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)100.8 672.8 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
685.8 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)100.8 698.8 Q(et loss.)
-.11 E([WBL04] proposes modifying the loss measurement process to accou\
nt for the)5.5 E(bias in f)100.8 711.8 Q -.22(avo)-.11 G 2.75(ro).22 G
2.75(fs)-2.75 G(maller pack)-2.75 E(ets.)-.11 E(Flo)72 769 Q(yd/K)-.11 E
276.229(ohler Section)-.385 F 2.75(A. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E(The TFRC-SP v)100.8 85 Q
(ariant of TFRC proposed in our document dif)-.275 E
(fers from [WBL04] in)-.275 E(restricting its attention to \215o)100.8
98 Q(ws that send at most 100 pack)-.275 E(ets per second.)-.11 E
(The TFRC-SP)5.5 E -.275(va)100.8 111 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)
100.8 124 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,)100.8 140.6 Q("unbiasing", "virtual pack)100.8
153.6 Q(ets", "random sampling", and "Loss Insensiti)-.11 E .33 -.165
(ve P)-.275 H(eriod \(LIP\)).165 E 2.75(scaling". [WBL04])100.8 166.6 R
(\214nds only the second and third methods suf)2.75 E(\214ciently rob)
-.275 E(ust when the)-.22 E(netw)100.8 179.6 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)100.8 192.6 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
205.6 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)100.8 218.6 Q
(xcept that randomization is not used; instead of)-.165 E
(randomization, the e)100.8 231.6 Q(xact pack)-.165 E
(et loss rate is computed for short loss interv)-.11 E(als, and the)
-.275 E(standard loss e)100.8 244.6 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)100.8 257.6 Q(arying)
-.275 E -.165(ove)100.8 270.6 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)100.8 283.6 Q(ws.)-.275 E
([WBL04] produces both a byte-mode and a pack)100.8 300.2 Q(et-mode v)
-.11 E(ariant of the TFRC transport)-.275 E(protocol, for connections o)
100.8 313.2 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 1.76 -.88(We w)100.8 326.2 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
339.2 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
352.2 Q(et-dropping beha)-.11 E(viors of)-.22 E
(the routers along the path.)100.8 365.2 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)100.8 378.2 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)100.8 391.2 Q
(-byte and per)-.22 E(-pack)-.22 E
(et dropping \(and on paths with multiple congested routers,)-.11 E
(some with per)100.8 404.2 Q(-byte dropping mechanisms, some with per)
-.22 E(-pack)-.22 E(et dropping mechanisms,)-.11 E
(and some with dropping mechanisms that lie some)100.8 417.2 Q
(where between per)-.275 E(-byte and per)-.22 E(-pack)-.22 E(et\).)-.11
E 2.75<418c>100.8 433.8 S(rst step will be to in)-2.75 E -.165(ve)-.44 G
(stig).165 E(ate the range of beha)-.055 E
(viors actually present in today')-.22 E(s)-.605 E(netw)100.8 446.8 Q
(orks, in terms of pack)-.11 E(et-dropping as a function of pack)-.11 E
(et size.)-.11 E 1.76 -.88(We w)5.5 H(ill report on these).88 E(in)100.8
459.8 Q -.165(ve)-.44 G(stig).165 E(ations in a separate document.)-.055
E/F1 11/Times-Bold@0 SF(B.)72 485.8 Q/F2 14/Times-Bold@0 SF -.35(Ap)5.5
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)100.8 502.4 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)100.8 515.4 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
528.4 Q(ge pack)-.198 E(ets are more lik)-.11 E
(ely to be dropped than small ones.)-.11 E(Flo)72 769 Q(yd/K)-.11 E
276.834(ohler Section)-.385 F 2.75(B. [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 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E/F1 11/Courier@0 SF
(Advantages of Packet Dropping Independent of Packet Size:)100.8 85 Q
(---------------------------------------------------------)100.8 98 Q
6.6(1. Adds)100.8 111 R
(another incentive for end nodes to use large packets.)6.6 E 6.6
(2. Matches)100.8 137 R
(an environment with a limitation in pps rather than)6.6 E(bps.)127.2
150 Q(---------------------------------------------------------)100.8
163 Q(Advantages of Packet Dropping as a Function of Packet Size:)100.8
189 Q(---------------------------------------------------------)100.8
202 Q 6.6(1. Small)100.8 215 R
(control packets are less likely to be dropped than are)6.6 E
(large data packets, improving TCP performance.)127.2 228 Q 6.6
(2. Matches)100.8 254 R
(an environment with a limitation in bps rather than)6.6 E(pps.)127.2
267 Q 6.6(3. Reduces)100.8 293 R
(the penalty of TCP and other transport protocols)6.6 E
(against flows with small packets \(where the allowed sending)127.2 306
Q(rate is roughly a linear function of packet size\).)127.2 319 Q 6.6
(4. A)100.8 345 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)127.2 358 Q
(worst-case queueing delay in seconds.)127.2 371 Q
(---------------------------------------------------------)100.8 384 Q
/F2 14/Times-Bold@0 SF(Normati)72 416 Q .28 -.14(ve R)-.14 H(efer).14 E
(ences)-.252 E F0([RFC 2119] S. Bradner)100.8 432.6 Q 2.75(.K)-.605 G
.33 -.165(ey W)-3.025 H(ords F)-.715 E
(or Use in RFCs to Indicate Requirement Le)-.165 E -.165(ve)-.275 G
(ls. RFC).165 E(2119.)129.6 445.6 Q([RFC 2434] T)100.8 462.2 Q 2.75(.N)
-.814 G(arten and H. Alv)-2.75 E 2.75(estrand. Guidelines)-.165 F
(for Writing an IAN)2.75 E 2.75(AC)-.385 G(onsiderations)-2.75 E
(Section in RFCs.)129.6 475.2 Q(RFC 2434.)5.5 E([RFC 3448] M. Handle)
100.8 491.8 Q 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 Control)-2.75 E(\(TFRC\): Protocol Speci\214cat\
ion, RFC 3448, Proposed Standard, January 2003.)129.6 504.8 Q F2(Inf)72
530.8 Q(ormati)-.35 E .28 -.14(ve R)-.14 H(efer).14 E(ences)-.252 E F0
([CCID 3 PR)100.8 547.4 Q(OFILE] S. Flo)-.44 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 DCCP Congestion)2.75 E
(Control ID 3: TFRC Congestion Control.)129.6 560.4 Q
(draft-ietf-dccp-ccid3-11.txt, w)5.5 E(ork in)-.11 E
(progress, March 2005.)129.6 573.4 Q([DCCP] E. K)100.8 590 Q(ohler)-.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 Protocol,)
2.75 E(draft-ietf-dccp-spec-13.txt, w)129.6 603 Q
(ork in progress, December 2005.)-.11 E([P04] T)100.8 619.6 Q 2.75(.P)
-.814 G(helan, TFRC with Self-Limiting Sources, October 2004.)-2.75 E
(URL)5.5 E("http://www)129.6 632.6 Q(.phelan-4.com/dccp/".)-.715 E
([RFC 3714] S. Flo)100.8 649.2 Q(yd and J. K)-.11 E(empf, Editors.)-.275
E(IAB Concerns Re)5.5 E -.055(ga)-.165 G(rding Congestion Control).055 E
(for V)129.6 662.2 Q(oice T)-1.419 E(raf)-.385 E(\214c in the Internet.)
-.275 E(RFC 3714.)5.5 E([V00] P)100.8 678.8 Q 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)5.5 E -.77(Te)
129.6 691.8 S(chnical Report TR-00-008, April 2000.).77 E(URL)5.5 E
("http://www)129.6 704.8 Q(.icsi.berk)-.715 E(ele)-.11 E -.715(y.)-.165
G(edu/techreports/2000.abstracts/ tr).715 E(-00-008.html".)-.22 E(Flo)72
769 Q(yd/K)-.11 E 328.171(ohler [P)-.385 F(age 25])-.165 E 0 Cg EP
%%Page: 26 26
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 79.523(-DRAFT Expires:)-1.012 F
(July 2006)2.75 E(January 2006)115.79 E([WBL04] J. W)100.8 85 Q(idmer)
-.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 Control for)5.5 E(Flo)129.6 98 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)72 124 S
(thors' Addr).7 E(esses)-.252 E F0(Sally Flo)100.8 140.6 Q(yd <\215o)
-.11 E(yd@icir)-.11 E(.or)-.605 E(g>)-.198 E
(ICSI Center for Internet Research)100.8 153.6 Q
(1947 Center Street, Suite 600)100.8 166.6 Q(Berk)100.8 179.6 Q(ele)-.11
E 1.43 -.715(y, C)-.165 H 2.75(A9).715 G(4704)-2.75 E(USA)100.8 192.6 Q
(Eddie K)100.8 209.2 Q(ohler <k)-.385 E(ohler@cs.ucla.edu>)-.11 E
(4531C Boelter Hall)100.8 222.2 Q(UCLA Computer Science Department)100.8
235.2 Q(Los Angeles, CA 90095)100.8 248.2 Q(USA)100.8 261.2 Q F1
(Full Copyright Statement)72 287.2 Q F0(Cop)100.8 303.8 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)100.8 316.8 Q
(xcept as set forth therein, the authors retain all)-.165 E
(their rights.)100.8 329.8 Q
(This document and the information contained herein are pro)100.8 346.4
Q(vided on an "AS IS" basis and)-.165 E(THE CONTRIB)100.8 359.4 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
372.4 Q(ENGINEERING T)100.8 385.4 Q(ASK FORCE DISCLAIM ALL W)-1.023 E
(ARRANTIES, EXPRESS OR)-1.32 E(IMPLIED, INCLUDING B)100.8 398.4 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)
100.8 411.4 Q(TION HEREIN WILL NO)-1.221 E 2.75(TI)-.44 G
(NFRINGE ANY RIGHTS OR ANY)-2.75 E(IMPLIED W)100.8 424.4 Q
(ARRANTIES OF MERCHANT)-1.32 E(ABILITY OR FITNESS FOR A)-1.023 E -1.012
(PA)100.8 437.4 S -.66(RT)1.012 G(ICULAR PURPOSE.).66 E F1
(Intellectual Pr)72 463.4 Q(operty)-.252 E F0(The IETF tak)100.8 480 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)100.8 493 Q
(technology described in this document or the e)100.8 506 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)100.8 519 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)100.8 532 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.)100.8 545 Q
(Copies of IPR disclosures made to the IETF Secretariat and an)100.8
561.6 Q 2.75(ya)-.165 G(ssurances of licenses to)-2.75 E(be made a)100.8
574.6 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)100.8 587.6 Q(speci\214cati\
on can be obtained from the IETF on-line IPR repository at)100.8 600.6 Q
(http://www)100.8 613.6 Q(.ietf.or)-.715 E(g/ipr)-.198 E(.)-.605 E
(The IETF in)100.8 630.2 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
643.2 Q -.165(ve)-.165 G 2.75(rt).165 G(echnology that may be)-2.75 E
(required to implement this standard.)100.8 656.2 Q
(Please address the information to the IETF at ietf-)5.5 E(ipr@ietf.or)
100.8 669.2 Q(g.)-.198 E(Flo)72 769 Q(yd/K)-.11 E 328.171(ohler [P)-.385
F(age 26])-.165 E 0 Cg EP
%%Trailer
end
%%EOF