One document matched: draft-irtf-tmrg-tools-04.ps

%!PS-Adobe-3.0
%%Creator: groff version 1.18.1
%%CreationDate: Sun Jul  8 21:08:28 2007
%%DocumentNeededResources: font Courier-Bold
%%+ font Times-Bold
%%+ font Times-Roman
%%DocumentSuppliedResources: procset grops 1.18 1
%%Pages: 22
%%PageOrder: Ascend
%%Orientation: Portrait
%%EndComments
%%BeginProlog
%%BeginResource: procset grops 1.18 1
/setpacking where{
pop
currentpacking
true setpacking
}if
/grops 120 dict dup begin
/SC 32 def
/A/show load def
/B{0 SC 3 -1 roll widthshow}bind def
/C{0 exch ashow}bind def
/D{0 exch 0 SC 5 2 roll awidthshow}bind def
/E{0 rmoveto show}bind def
/F{0 rmoveto 0 SC 3 -1 roll widthshow}bind def
/G{0 rmoveto 0 exch ashow}bind def
/H{0 rmoveto 0 exch 0 SC 5 2 roll awidthshow}bind def
/I{0 exch rmoveto show}bind def
/J{0 exch rmoveto 0 SC 3 -1 roll widthshow}bind def
/K{0 exch rmoveto 0 exch ashow}bind def
/L{0 exch rmoveto 0 exch 0 SC 5 2 roll awidthshow}bind def
/M{rmoveto show}bind def
/N{rmoveto 0 SC 3 -1 roll widthshow}bind def
/O{rmoveto 0 exch ashow}bind def
/P{rmoveto 0 exch 0 SC 5 2 roll awidthshow}bind def
/Q{moveto show}bind def
/R{moveto 0 SC 3 -1 roll widthshow}bind def
/S{moveto 0 exch ashow}bind def
/T{moveto 0 exch 0 SC 5 2 roll awidthshow}bind def
/SF{
findfont exch
[exch dup 0 exch 0 exch neg 0 0]makefont
dup setfont
[exch/setfont cvx]cvx bind def
}bind def
/MF{
findfont
[5 2 roll
0 3 1 roll
neg 0 0]makefont
dup setfont
[exch/setfont cvx]cvx bind def
}bind def
/level0 0 def
/RES 0 def
/PL 0 def
/LS 0 def
/MANUAL{
statusdict begin/manualfeed true store end
}bind def
/PLG{
gsave newpath clippath pathbbox grestore
exch pop add exch pop
}bind def
/BP{
/level0 save def
1 setlinecap
1 setlinejoin
72 RES div dup scale
LS{
90 rotate
}{
0 PL translate
}ifelse
1 -1 scale
}bind def
/EP{
level0 restore
showpage
}bind def
/DA{
newpath arcn stroke
}bind def
/SN{
transform
.25 sub exch .25 sub exch
round .25 add exch round .25 add exch
itransform
}bind def
/DL{
SN
moveto
SN
lineto stroke
}bind def
/DC{
newpath 0 360 arc closepath
}bind def
/TM matrix def
/DE{
TM currentmatrix pop
translate scale newpath 0 0 .5 0 360 arc closepath
TM setmatrix
}bind def
/RC/rcurveto load def
/RL/rlineto load def
/ST/stroke load def
/MT/moveto load def
/CL/closepath load def
/Fr{
setrgbcolor fill
}bind def
/Fk{
setcmykcolor fill
}bind def
/Fg{
setgray fill
}bind def
/FL/fill load def
/LW/setlinewidth load def
/Cr/setrgbcolor load def
/Ck/setcmykcolor load def
/Cg/setgray load def
/RE{
findfont
dup maxlength 1 index/FontName known not{1 add}if dict begin
{
1 index/FID ne{def}{pop pop}ifelse
}forall
/Encoding exch def
dup/FontName exch def
currentdict end definefont pop
}bind def
/DEFS 0 def
/EBEGIN{
moveto
DEFS begin
}bind def
/EEND/end load def
/CNT 0 def
/level1 0 def
/PBEGIN{
/level1 save def
translate
div 3 1 roll div exch scale
neg exch neg exch translate
0 setgray
0 setlinecap
1 setlinewidth
0 setlinejoin
10 setmiterlimit
[]0 setdash
/setstrokeadjust where{
pop
false setstrokeadjust
}if
/setoverprint where{
pop
false setoverprint
}if
newpath
/CNT countdictstack def
userdict begin
/showpage{}def
}bind def
/PEND{
clear
countdictstack CNT sub{end}repeat
level1 restore
}bind def
end def
/setpacking where{
pop
setpacking
}if
%%EndResource
%%IncludeResource: font Courier-Bold
%%IncludeResource: font Times-Bold
%%IncludeResource: font Times-Roman
grops begin/DEFS 1 dict def DEFS begin/u{.001 mul}bind def end/RES 72
def/PL 841.89 def/LS false def/ENC0[/asciicircum/asciitilde/Scaron
/Zcaron/scaron/zcaron/Ydieresis/trademark/quotesingle/Euro/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/space/exclam/quotedbl/numbersign/dollar/percent
/ampersand/quoteright/parenleft/parenright/asterisk/plus/comma/hyphen
/period/slash/zero/one/two/three/four/five/six/seven/eight/nine/colon
/semicolon/less/equal/greater/question/at/A/B/C/D/E/F/G/H/I/J/K/L/M/N/O
/P/Q/R/S/T/U/V/W/X/Y/Z/bracketleft/backslash/bracketright/circumflex
/underscore/quoteleft/a/b/c/d/e/f/g/h/i/j/k/l/m/n/o/p/q/r/s/t/u/v/w/x/y
/z/braceleft/bar/braceright/tilde/.notdef/quotesinglbase/guillemotleft
/guillemotright/bullet/florin/fraction/perthousand/dagger/daggerdbl
/endash/emdash/ff/fi/fl/ffi/ffl/dotlessi/dotlessj/grave/hungarumlaut
/dotaccent/breve/caron/ring/ogonek/quotedblleft/quotedblright/oe/lslash
/quotedblbase/OE/Lslash/.notdef/exclamdown/cent/sterling/currency/yen
/brokenbar/section/dieresis/copyright/ordfeminine/guilsinglleft
/logicalnot/minus/registered/macron/degree/plusminus/twosuperior
/threesuperior/acute/mu/paragraph/periodcentered/cedilla/onesuperior
/ordmasculine/guilsinglright/onequarter/onehalf/threequarters
/questiondown/Agrave/Aacute/Acircumflex/Atilde/Adieresis/Aring/AE
/Ccedilla/Egrave/Eacute/Ecircumflex/Edieresis/Igrave/Iacute/Icircumflex
/Idieresis/Eth/Ntilde/Ograve/Oacute/Ocircumflex/Otilde/Odieresis
/multiply/Oslash/Ugrave/Uacute/Ucircumflex/Udieresis/Yacute/Thorn
/germandbls/agrave/aacute/acircumflex/atilde/adieresis/aring/ae/ccedilla
/egrave/eacute/ecircumflex/edieresis/igrave/iacute/icircumflex/idieresis
/eth/ntilde/ograve/oacute/ocircumflex/otilde/odieresis/divide/oslash
/ugrave/uacute/ucircumflex/udieresis/yacute/thorn/ydieresis]def
/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
(S. Floyd)197.998 E 287.998(INTERNET-DRAFT E.)72 98 R(Kohler)6 E
(Intended status: Informational)72 111 Q(Editors)209.998 E
(Expires: January 2008)72 124 Q 6(8J)239.998 G(uly 2007)-6 E/F1 14
/Times-Bold@0 SF -1.288(To)105.958 149 S(ols f)1.288 E(or the Ev)-.35 E
(aluation of Simulation and T)-.14 E(estbed Scenarios)-1.288 E
(draft-irtf-tmr)205.5 162 Q(g-tools-04.ps)-.14 E/F2 11/Times-Bold@0 SF
(Status of this Memo)72 194 Q/F3 11/Times-Roman@0 SF
(By submitting this Internet-Draft, each author represents that an)100.8
210.6 Q 2.75(ya)-.165 G(pplicable patent or other)-2.75 E
(IPR claims of which he or she is a)100.8 223.6 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)100.8 236.6
Q -.11(wa)-.165 G
(re will be disclosed, in accordance with Section 6 of BCP 79.).11 E
(Internet-Drafts are w)100.8 253.2 Q
(orking documents of the Internet Engineering T)-.11 E(ask F)-.88 E
(orce \(IETF\), its)-.165 E(areas, and its w)100.8 266.2 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.)100.8 279.2 Q
(Internet-Drafts are draft documents v)100.8 295.8 Q
(alid for a maximum of six months and may be)-.275 E
(updated, replaced, or obsoleted by other documents at an)100.8 308.8 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)
100.8 321.8 Q(ork in progress.")-.11 E
(The list of current Internet-Drafts can be accessed at http://www)100.8
338.4 Q(.ietf.or)-.715 E(g/ietf/1id-)-.198 E(abstracts.txt.)100.8 351.4
Q(The list of Internet-Draft Shado)100.8 368 Q 2.75(wD)-.275 G
(irectories can be accessed at)-2.75 E(http://www)100.8 381 Q(.ietf.or)
-.715 E(g/shado)-.198 E -.715(w.)-.275 G(html.).715 E
(This Internet-Draft will e)100.8 397.6 Q(xpire on January 2008.)-.165 E
F2(Abstract)72 423.6 Q F3(This document describes tools for the e)100.8
440.2 Q -.275(va)-.275 G
(luation of simulation and testbed scenarios used).275 E
(in research on Internet congestion control mechanisms.)100.8 453.2 Q
1.76 -.88(We b)5.5 H(elie).88 E .33 -.165(ve t)-.275 H(hat research in)
.165 E(congestion control mechanisms has been seriously hampered by the\
 lack of good models)100.8 466.2 Q
(underpinning analysis, simulation, and testbed e)100.8 479.2 Q
(xperiments, and that tools for the)-.165 E -.275(eva)100.8 492.2 S(lua\
tion of simulation and testbed scenarios can help in the construction o\
f better).275 E(scenarios, based on better underlying models.)100.8
505.2 Q(One use of the tools described in this)5.5 E
(document is in comparing k)100.8 518.2 Q .33 -.165(ey c)-.11 H
(haracteristics of test scenarios with kno).165 E(wn characteristics)
-.275 E(from the di)100.8 531.2 Q -.165(ve)-.275 G(rse and e).165 E
-.165(ve)-.275 G -.22(r-).165 G(changing real w).22 E 2.75(orld. T)-.11
F(ools characterizing the aggre)-.88 E -.055(ga)-.165 G(te traf).055 E
<8c63>-.275 E(on a link include the distrib)100.8 544.2 Q(ution of per)
-.22 E(-pack)-.22 E(et round-trip times, the distrib)-.11 E(ution of)
-.22 E(connection sizes, and the lik)100.8 557.2 Q 2.75(e. T)-.11 F
(ools characterizing end-to-end paths include drop rates as)-.88 E 2.75
(af)100.8 570.2 S(unction of pack)-2.75 E(et size and of b)-.11 E
(urst size, the synchronization ratio between tw)-.22 E 2.75(oe)-.11 G
(nd-to-)-2.75 E(end TCP \215o)100.8 583.2 Q(ws, and the lik)-.275 E 2.75
(e. F)-.11 F(or each characteristic, we describe what aspects of the)
-.165 E(scenario determine this characteristic, ho)100.8 596.2 Q 2.75
(wt)-.275 G(he characteristic can af)-2.75 E(fect the results of)-.275 E
(simulations and e)100.8 609.2 Q(xperiments for the e)-.165 E -.275(va)
-.275 G(luation of congestion control mechanisms, and).275 E
(what is kno)100.8 622.2 Q(wn about this characteristic in the real w)
-.275 E 2.75(orld. W)-.11 F 2.75(ea)-.88 G(lso e)-2.75 E(xplain wh)-.165
E 2.75(yt)-.055 G(he use of)-2.75 E(such tools can add considerable po)
100.8 635.2 Q(wer to our understanding and e)-.275 E -.275(va)-.275 G
(luation of simulation).275 E(and testbed scenarios.)100.8 648.2 Q(Flo)
72 769 Q(yd, K)-.11 E 331.229(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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 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. Introduction)100.8 110 Q F0 11(.......................)11.01 G F2(3)
11.5 E(2. T)100.8 122 Q(ools)-.8 E F0 11(.........................)10.97
G F2(4)11.5 E(2.1. Characterizing Aggre)115.8 134 Q -.05(ga)-.15 G(te T)
.05 E(raf)-.35 E(\214c on a Link)-.25 E F0 11(.............)2.69 G F2(4)
11.5 E(2.2. Characterizing an End-to-End P)115.8 146 Q(ath)-.15 E F0 11
(...............)7.85 G F2(4)11.5 E(2.3. Other Characteristics)115.8 158
Q F0 11(...................)8.25 G F2(4)11.5 E(3. The Distrib)100.8 170
Q(ution of Per)-.2 E(-pack)-.2 E(et Round-trip T)-.1 E(imes)-.35 E F0 11
(.............)2.43 G F2(4)11.5 E(4. The Distrib)100.8 182 Q
(ution of Connection Sizes)-.2 E F0 11(................)7.46 G F2(5)11.5
E(5. The Distrib)100.8 194 Q(ution of P)-.2 E(ack)-.15 E(et Sizes)-.1 E
F0 11(.................)13.41 G F2(6)11.5 E(6. The Ratio Between F)100.8
206 Q(orw)-.15 E(ard-path and Re)-.1 E -.15(ve)-.25 G(rse-path T).15 E
(raf)-.35 E<8c63>-.25 E F0 11(..........)4.4 G F2(6)11.5 E
(7. The Distrib)100.8 218 Q(ution of Per)-.2 E(-P)-.2 E(ack)-.15 E
(et Peak Flo)-.1 E 2.5(wR)-.25 G(ates)-2.5 E F0 11(.............)6.09 G
F2(7)11.5 E(8. The Distrib)100.8 230 Q(ution of T)-.2 E
(ransport Protocols.)-.35 E F0 11(...............)9.62 G F2(7)11.5 E
(9. The Synchronization Ratio)100.8 242 Q F0 11(...................)7.68
G F2(8)11.5 E(10. Drop or Mark Rates as a Function of P)100.8 254 Q(ack)
-.15 E(et Size)-.1 E F0 11(............)11.15 G F2(9)11.5 E
(11. Drop Rates as a Function of Burst Size.)100.8 266 Q F0 11
(...............)7.13 G F2(10)6.5 E
(12. Drop Rates as a Function of Sending Rate.)100.8 278 Q F0 11
(..............)8.66 G F2(11)6.5 E
(13. Congestion Control Mechanisms for T)100.8 290 Q(raf)-.35 E
(\214c, along with Sender and Recei)-.25 E -.15(ve)-.25 G 2.5(rB).15 G
(uf)-2.5 E(fer Sizes.)-.25 E F0(.)-1.9 E F2(12)15.6 E
(14. Characterization of Congested Links in T)100.8 302 Q
(erms of Bandwidth and T)-.7 E(ypical Le)-.8 E -.15(ve)-.25 G
(ls of Congestion).15 E F0(.)-2.08 E F2(12)8.92 E(14.1. Bandwidth)115.8
314 Q F0 11(......................)2.81 G F2(12)6.5 E
(14.2. Queue Management Mechanisms)115.8 326 Q F0 11(...............)
9.09 G F2(12)6.5 E(14.3. T)115.8 338 Q(ypical Le)-.8 E -.15(ve)-.25 G
(ls of Congestion).15 E F0 11(................)11.8 G F2(12)6.5 E
(15. Characterization of Challenging Lo)100.8 350 Q(wer Layers.)-.25 E
F0 11(.............)4.08 G F2(12)6.5 E(15.1. Error Losses)115.8 362 Q F0
11(.....................)9.63 G F2(12)6.5 E(15.2. P)115.8 374 Q(ack)-.15
E(et Reordering)-.1 E F0 11(....................).3 G F2(13)6.5 E
(15.3. Delay V)115.8 386 Q(ariation)-1.11 E F0 11(....................)
11.16 G F2(14)6.5 E(15.4. Bandwidth V)115.8 398 Q(ariation)-1.11 E F0 11
(...................)4.9 G F2(14)6.5 E(15.5. Bandwidth and Latenc)115.8
410 Q 2.5(yA)-.15 G(symmetry)-2.5 E F0 11(..............)11.59 G F2(15)
6.5 E(15.6. Queue Management Mechanisms)115.8 422 Q F0 11
(...............)9.09 G F2(16)6.5 E(16. Netw)100.8 434 Q(ork Changes Af)
-.1 E(fecting Congestion)-.25 E F0 11(...............)4.99 G F2(16)6.5 E
(16.1. Routing Changes:)115.8 446 Q(Routing Loops)5 E F0 11
(...............)6.27 G F2(16)6.5 E(16.2. Routing Changes:)115.8 458 Q
(Fluttering)5 E F0 11(................)12.8 G F2(16)6.5 E
(16.3. Routing Changes:)115.8 470 Q(Routing Asymmetry)5 E F0 11
(.............)11.55 G F2(17)6.5 E
(16.4. Link Disconnections and Intermittent Link Connecti)115.8 482 Q
(vity)-.25 E F0 11(.........).43 G F2(17)6.5 E(16.5. Changes in W)115.8
494 Q(ireless Links: Mobility)-.4 E F0 11(..............)9.32 G F2(18)
6.5 E(17. Using the T)100.8 506 Q(ools Presented in this Document)-.8 E
F0 11(..............)3.34 G F2(18)6.5 E(18. Related W)100.8 518 Q(ork)
-.8 E F0 11(.......................).43 G F2(18)6.5 E(19. Conclusions)
100.8 530 Q F0 11(.......................)6 G F2(18)6.5 E
(20. Security Considerations)100.8 542 Q F0 11(....................).87
G F2(18)6.5 E(21. IAN)100.8 554 Q 2.5(AC)-.35 G(onsiderations)-2.5 E F0
11(....................)9.56 G F2(19)6.5 E(22. Ackno)100.8 566 Q
(wledgements)-.25 E F0 11(.....................)3.77 G F2(19)6.5 E
(Informati)100.8 578 Q .3 -.15(ve R)-.25 H(eferences).15 E F0 11
(.....................)4.22 G F2(19)6.5 E(Editors' Addresses)100.8 590 Q
F0 11(......................)8.38 G F2(21)6.5 E(Full Cop)100.8 602 Q
(yright Statement)-.1 E F0 11(....................)10.12 G F2(22)6.5 E
(Intellectual Property)100.8 614 Q F0 11(......................)2.27 G
F2(22)6.5 E F0(Flo)72 769 Q(yd, K)-.11 E 331.229(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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 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-irtf-tmr)103.55 111 Q(g-tools-03.txt:)-.198
E 2.75(*N)103.55 124 S 2.75(oc)-2.75 G(hanges.)-2.75 E
(Changes from draft-irtf-tmr)103.55 150 Q(g-tools-02.txt:)-.198 E 2.75
(*A)103.55 163 S(dded sections on Challenging Lo)-2.75 E
(wer Layers and Netw)-.275 E(ork)-.11 E(Changes af)109.05 176 Q
(fecting Congestion.)-.275 E(Contrib)5.5 E(uted by Jasani Rohan,)-.22 E
(with Julie T)109.05 189 Q(arr)-.88 E 2.75(,T)-.44 G(on)-3.63 E 2.75(yD)
-.165 G(esimone, Christou Christos, and)-2.75 E -1.221(Ve)109.05 202 S
(mulapalli Archana.)1.221 E 2.75(*M)103.55 228 S(inor editing.)-2.75 E
(Changes from draft-irtf-tmr)103.55 254 Q(g-tools-01.txt:)-.198 E 2.75
(*A)103.55 267 S
(dded section on "Drop Rates as a Function of Sending Rate.")-2.75 E
2.75(*A)103.55 293 S(dded a number of ne)-2.75 E 2.75(wr)-.275 G
(eferences.)-2.75 E(END OF SECTION T)100.8 319 Q 2.75(OB)-.198 G 2.75
(ED)-2.75 G(ELETED.)-2.75 E/F1 11/Times-Bold@0 SF(1.)72 358 Q/F2 14
/Times-Bold@0 SF(Intr)5.5 E(oduction)-.252 E F0
(This document discusses tools for the e)100.8 374.6 Q -.275(va)-.275 G
(luation of simulation and testbed scenarios used).275 E
(in research on Internet congestion control mechanisms.)100.8 387.6 Q
(These tools include b)5.5 E(ut are not)-.22 E(limited to measurement t\
ools; the tools discussed in this document are lar)100.8 400.6 Q(gely w)
-.198 E(ays of)-.11 E(characterizing aggre)100.8 413.6 Q -.055(ga)-.165
G(te traf).055 E
(\214c on a link, or characterizing the end-to-end path.)-.275 E
(One use of)5.5 E(these tools is for understanding k)100.8 426.6 Q .33
-.165(ey c)-.11 H(haracteristics of test scenarios; man).165 E 2.75(yc)
-.165 G(haracteristics,)-2.75 E(such as the distrib)100.8 439.6 Q
(ution of per)-.22 E(-pack)-.22 E(et round-trip times on the link, don')
-.11 E 2.75(tc)-.198 G(ome from a single)-2.75 E(input parameter b)100.8
452.6 Q(ut are determined by a range of inputs.)-.22 E 2.75(As)5.5 G
(econd use of the tools is to)-2.75 E(compare k)100.8 465.6 Q .33 -.165
(ey c)-.11 H(haracteristics of test scenarios with what is kno).165 E
(wn of the same)-.275 E(characteristics of the past and current Interne\
t, and with what can be conjectured about)100.8 478.6 Q
(these characteristics of future netw)100.8 491.6 Q 2.75(orks. This)-.11
F(paper follo)2.75 E(ws the general approach from)-.275 E
("Internet Research Needs Better Models" [FK02].)100.8 504.6 Q(As an e)
100.8 521.2 Q(xample of the po)-.165 E
(wer of tools for characterizing scenarios, a great deal is kno)-.275 E
(wn)-.275 E(about the distrib)100.8 534.2 Q
(ution of connection sizes on a link, or equi)-.22 E -.275(va)-.275 G
(lently).275 E 2.75(,t)-.715 G(he distrib)-2.75 E(ution of per)-.22 E(-)
-.22 E(pack)100.8 547.2 Q(et sequence numbers.)-.11 E
(It has been conjectured that a hea)5.5 E(vy-tailed distrib)-.22 E
(ution of)-.22 E(connection sizes is an in)100.8 560.2 Q -.275(va)-.44 G
(riant feature of Internet traf).275 E 2.75(\214c. A)-.275 F
(test scenario with mostly long-)2.75 E(li)100.8 573.2 Q -.165(ve)-.275
G 2.75(dt).165 G(raf)-2.75 E(\214c, or with a mix with only long-li)
-.275 E -.165(ve)-.275 G 2.75(da).165 G(nd v)-2.75 E(ery short \215o)
-.165 E(ws, does not ha)-.275 E .33 -.165(ve a)-.22 H(realistic distrib)
100.8 586.2 Q(ution of connection sizes, and can gi)-.22 E .33 -.165
(ve u)-.275 H(nrealistic results in simulations or).165 E -.165(ex)100.8
599.2 S(periments e).165 E -.275(va)-.275 G
(luating congestion control mechanisms.).275 E -.165(Fo)5.5 G 2.75(ri)
.165 G(nstance, the distrib)-2.75 E(ution of)-.22 E
(connection sizes mak)100.8 612.2 Q(es clear the fraction of traf)-.11 E
(\214c on a link from medium-sized)-.275 E(connections, e.g., with pack)
100.8 625.2 Q(et sequence numbers from 100 to 1000.)-.11 E
(These medium-sized)5.5 E(connections can slo)100.8 638.2 Q
(w-start up to a lar)-.275 E(ge congestion windo)-.198 E 1.43 -.715
(w, p)-.275 H(ossibly coming to an abrupt).715 E(stop soon afterw)100.8
651.2 Q(ards, contrib)-.11 E(uting signi\214cantly to the b)-.22 E
(urstiness of the aggre)-.22 E -.055(ga)-.165 G(te traf).055 E(\214c,)
-.275 E(and to the problems f)100.8 664.2 Q(acing congestion control.)
-.11 E(In the sections belo)100.8 680.8 Q 2.75(ww)-.275 G 2.75(ew)-2.75
G(ill discuss a number of tools for describing and e)-2.75 E -.275(va)
-.275 G(luating).275 E(scenarios, sho)100.8 693.8 Q 2.75(wh)-.275 G .55
-.275(ow t)-2.75 H(hese characteristics can af).275 E
(fect the results of research on congestion)-.275 E
(control mechanisms, and summarize what is kno)100.8 706.8 Q
(wn about these characteristics in real-)-.275 E -.11(wo)100.8 719.8 S
(rld netw).11 E(orks.)-.11 E(Flo)72 769 Q(yd, K)-.11 E 281.729
(ohler Section)-.385 F 2.75(1. [P)2.75 F(age 3])-.165 E 0 Cg EP
%%Page: 4 4
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E/F1 11/Times-Bold@0 SF(2.)72 85
Q/F2 14/Times-Bold@0 SF -1.288(To)5.5 G(ols)1.288 E F0
(The tools or characteristics that we discuss are the follo)100.8 101.6
Q(wing.)-.275 E F1(2.1.)72 127.6 Q/F3 13/Times-Bold@0 SF
(Characterizing Aggr)5.5 E(egate T)-.234 E(raf\214c on a Link)-.962 E F0
12.65<8344>100.8 144.2 S(istrib)-12.65 E(ution of per)-.22 E(-pack)-.22
E(et round-trip times.)-.11 E 12.65<8344>100.8 160.8 S(istrib)-12.65 E
(ution of connection sizes.)-.22 E 12.65<8344>100.8 177.4 S(istrib)
-12.65 E(ution of pack)-.22 E(et sizes.)-.11 E 12.65<8352>100.8 194 S
(atio between forw)-12.65 E(ard-path and re)-.11 E -.165(ve)-.275 G
(rse-path traf).165 E(\214c.)-.275 E 12.65<8344>100.8 210.6 S(istrib)
-12.65 E(ution of peak \215o)-.22 E 2.75(wr)-.275 G(ates.)-2.75 E 12.65
<8344>100.8 227.2 S(istrib)-12.65 E(ution of transport protocols.)-.22 E
F1(2.2.)72 253.2 Q F3(Characterizing an End-to-End P)5.5 E(ath)-.13 E F0
12.65<8353>100.8 269.8 S(ynchronization ratio.)-12.65 E 12.65<8344>100.8
286.4 S(rop rates as a function of pack)-12.65 E(et size.)-.11 E 12.65
<8344>100.8 303 S(rop rates as a function of b)-12.65 E(urst size.)-.22
E 12.65<8344>100.8 319.6 S(rop rates as a function of sending rate.)
-12.65 E 12.65<8344>100.8 336.2 S -.165(eg)-12.65 G(ree of pack).165 E
(et drops.)-.11 E 12.65<8352>100.8 352.8 S(ange of queueing delay)-12.65
E(.)-.715 E F1(2.3.)72 378.8 Q F3(Other Characteristics)5.5 E F0 12.65
<8343>100.8 395.4 S(ongestion control mechanisms for traf)-12.65 E
(\214c, along with sender and recei)-.275 E -.165(ve)-.275 G 2.75(rb)
.165 G(uf)-2.97 E(fer sizes.)-.275 E 12.65<8343>100.8 412 S(haracteriza\
tion of congested links in terms of bandwidth and typical le)-12.65 E
-.165(ve)-.275 G(ls of).165 E(congestion \(in terms of pack)117.3 425 Q
(et drop rates\).)-.11 E 12.65<8343>100.8 441.6 S
(haracterization of congested links in terms of b)-12.65 E(uf)-.22 E
(fer size.)-.275 E 12.65<8343>100.8 458.2 S
(haracterization of challenging lo)-12.65 E
(wer layers in terms of reordering, delay v)-.275 E(ariation,)-.275 E
(pack)117.3 471.2 Q(et corruption, and the lik)-.11 E(e.)-.11 E 12.65
<8343>100.8 487.8 S(haracterization of netw)-12.65 E(ork changes af)-.11
E(fecting congestion, such as routing changes or)-.275 E(link outages.)
117.3 500.8 Q(Belo)100.8 517.4 Q 2.75(ww)-.275 G 2.75(ew)-2.75 G
(ill discuss each characteristic in turn, gi)-2.75 E
(ving the de\214nition, the f)-.275 E(actors)-.11 E
(determining that characteristic, the ef)100.8 530.4 Q
(fect on congestion control metrics, and what is kno)-.275 E(wn)-.275 E
(so f)100.8 543.4 Q(ar from measurement studies in the Internet.)-.11 E
F1(3.)72 569.4 Q F2(The Distrib)5.5 E(ution of P)-.28 E(er)-.28 E(-pack)
-.518 E(et Round-trip T)-.14 E(imes)-.252 E F0
(De\214nition: The distrib)100.8 586 Q(ution of per)-.22 E(-pack)-.22 E
(et round-trip times on a link is de\214ned formally by)-.11 E
(assigning to each pack)100.8 599 Q
(et the most recent round trip time measured for that end-to-end)-.11 E
2.75(connection. In)100.8 612 R
(practice, coarse-grained information is generally suf)2.75 E
(\214cient, e)-.275 E -.165(ve)-.275 G 2.75(nt).165 G(hough it)-2.75 E
(has been sho)100.8 625 Q(wn that there is signi\214cant v)-.275 E
(ariability in round-trip times within a TCP)-.275 E
(connection [AKSJ03], and it is suf)100.8 638 Q
(\214cient to assign to each pack)-.275 E
(et the \214rst round-trip time)-.11 E(measurement for that connection,\
 or to assign the current round-trip time estimate)100.8 651 Q
(maintained by the TCP connection.)100.8 664 Q(Determining f)100.8 693.6
Q(actors: The distrib)-.11 E(ution of per)-.22 E(-pack)-.22 E
(et round-trip times on a link is determined)-.11 E
(by end-to-end propag)100.8 706.6 Q
(ation delays, by queueing delays along end-to-end paths, and by the)
-.055 E(congestion control mechanisms used by the traf)100.8 719.6 Q
2.75(\214c. F)-.275 F(or e)-.165 E(xample, for a scenario using TCP)
-.165 E(,)-1.221 E(Flo)72 769 Q(yd, K)-.11 E 281.729(ohler Section)-.385
F 2.75(3. [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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E
(TCP connections with smaller round-trip times will recei)100.8 85 Q .33
-.165(ve a p)-.275 H(roportionally lar).165 E(ger fraction)-.198 E
(of traf)100.8 98 Q(\214c than competing TCP connections with lar)-.275
E(ger round-trip times, all else being)-.198 E
(equal, due to the dynamics of TCP f)100.8 111 Q -.22(avo)-.11 G
(ring \215o).22 E(ws with smaller round-trip times.)-.275 E(This will)
5.5 E(generally shift the distrib)100.8 124 Q(ution of per)-.22 E(-pack)
-.22 E(et R)-.11 E(TTs lo)-.66 E(wer relati)-.275 E .33 -.165(ve t)-.275
H 2.75(ot).165 G(he distrib)-2.75 E(ution of per)-.22 E(-)-.22 E
(connection R)100.8 137 Q(TTs, since short-R)-.66 E
(TT connections will ha)-.66 E .33 -.165(ve m)-.22 H(ore pack).165 E
(ets.)-.11 E(Ef)100.8 153.6 Q
(fect on congestion control metrics: The distrib)-.275 E(ution of per)
-.22 E(-pack)-.22 E(et round-trip times on a)-.11 E(link af)100.8 166.6
Q(fects the b)-.275 E(urstiness of the aggre)-.22 E -.055(ga)-.165 G
(te traf).055 E(\214c, and therefore can af)-.275 E(fect congestion)
-.275 E(control performance in a range of areas such as delay/throughpu\
t tradeof)100.8 179.6 Q 2.75(fs. The)-.275 F(distrib)100.8 192.6 Q
(ution of per)-.22 E(-pack)-.22 E(et round-trip times can also af)-.11 E
(fect metrics of f)-.275 E(airness, de)-.11 E(gree of)-.165 E
(oscillations, and the lik)100.8 205.6 Q 2.75(e. F)-.11 F(or e)-.165 E
(xample, long-term oscillations of queueing delay are more)-.165 E(lik)
100.8 218.6 Q(ely to occur in scenarios with a narro)-.11 E 2.75(wr)
-.275 G(ange of round-trip times [FK02].)-2.75 E
(Measurements: The distrib)100.8 235.2 Q(ution of per)-.22 E(-pack)-.22
E(et round-trip times for TCP traf)-.11 E(\214c on a link can)-.275 E
(be measured from a pack)100.8 248.2 Q(et trace with the passi)-.11 E
.33 -.165(ve T)-.275 H(CP round-trip time estimator from Jiang).165 E
(and Do)100.8 261.2 Q(vrolis [JD02].)-.165 E
([Add pointers to other estimators, such as ones mentioned in JD02.)5.5
E(Add a pointer to Mark Allman')100.8 274.2 Q 2.75(sl)-.605 G
(oss detection tool.])-2.75 E(Their paper sho)5.5 E(ws the distrib)-.275
E(ution of)-.22 E(per)100.8 287.2 Q(-pack)-.22 E
(et round-trip times for TCP pack)-.11 E(ets for a number of dif)-.11 E
(ferent links.)-.275 E -.165(Fo)5.5 G 2.75(rt).165 G(he links)-2.75 E
(measured, the percent of pack)100.8 300.2 Q
(ets with round-trip times at most 100 ms ranged from 30% to)-.11 E
(80%, and the percent of pack)100.8 313.2 Q
(ets with round-trip times at most 200 ms ranged from 55% to)-.11 E
(90%, depending on the link.)100.8 326.2 Q(In the NS simulator)100.8
342.8 Q 2.75(,t)-.44 G(he distrib)-2.75 E(ution of per)-.22 E(-pack)-.22
E(et round-trip times for TCP pack)-.11 E(ets on a)-.11 E
(link can be reported by the queue monitor)100.8 355.8 Q 2.75(,u)-.44 G
(sing TCP')-2.75 E 2.75(se)-.605 G(stimated round-trip time added to)
-2.75 E(pack)100.8 368.8 Q(et headers.)-.11 E
(This is illustrated in the v)5.5 E
(alidation test "./test-all-simple stats3" in the)-.275 E
(directory tcl/test.)100.8 381.8 Q(Scenarios: [FK02] sho)100.8 398.4 Q
(ws a relati)-.275 E -.165(ve)-.275 G
(ly simple scenario, with a dumbbell topology with four).165 E
(access links on each end, that gi)100.8 411.4 Q -.165(ve)-.275 G 2.75
(saf).165 G(airly realistic range of round-trip times.)-2.86 E
([Look for)5.5 E(the other citations to add.])100.8 424.4 Q/F1 11
/Times-Bold@0 SF(4.)72 450.4 Q/F2 14/Times-Bold@0 SF(The Distrib)5.5 E
(ution of Connection Sizes)-.28 E F0(De\214nition: Instead of the conne\
ction-based measurement of the distrib)100.8 467 Q(ution of connection)
-.22 E(sizes \(the total number of bytes or of data pack)100.8 480 Q
(ets in a connection\), we consider the pack)-.11 E(et-)-.11 E
(based measurement of the distrib)100.8 493 Q(ution of pack)-.22 E
(et sequence numbers.)-.11 E(The distrib)5.5 E(ution of)-.22 E(pack)
100.8 506 Q(et sequence numbers on a link is de\214ned by gi)-.11 E
(ving each pack)-.275 E(et a sequence number)-.11 E(,)-.44 E
(where the \214rst pack)100.8 519 Q
(et in a connection has sequence number 1, the second pack)-.11 E
(et has)-.11 E(sequence number 2, and so on.)100.8 532 Q(The distrib)5.5
E(ution of pack)-.22 E(et sequence numbers can be)-.11 E(deri)100.8 545
Q -.165(ve)-.275 G 2.75(di).165 G 2.75(nas)-2.75 G(traightforw)-2.75 E
(ard manner from the distrib)-.11 E(ution of connection sizes, and vice)
-.22 E -.165(ve)100.8 558 S 2.75(rsa; ho).165 F(we)-.275 E -.165(ve)
-.275 G .88 -.44(r, t).165 H(he distrib).44 E
(ution of connection sizes is more suited for traf)-.22 E
(\214c generators,)-.275 E(and the distrib)100.8 571 Q(ution of pack)
-.22 E(et sequence numbers is more suited for measuring and)-.11 E
(illustrating the pack)100.8 584 Q(ets actually seen on a link o)-.11 E
-.165(ve)-.165 G 2.75(ra\214).165 G -.165(xe)-2.75 G 2.75(di).165 G
(nterv)-2.75 E(al of time.)-.275 E(There has been)5.5 E 2.75(ac)100.8
597 S(onsiderably body of research o)-2.75 E -.165(ve)-.165 G 2.75(rt)
.165 G(he last ten years on the hea)-2.75 E(vy-tailed distrib)-.22 E
(ution of)-.22 E(connection sizes for traf)100.8 610 Q
(\214c on the Internet.)-.275 E([CBC95] [Add citations.])5.5 E
(Determining f)100.8 626.6 Q(actors: The distrib)-.11 E
(ution of connection sizes is lar)-.22 E(gely determined by the)-.198 E
(traf)100.8 639.6 Q(\214c generators used in a scenario.)-.275 E -.165
(Fo)5.5 G 2.75(re).165 G(xample, is there a single traf)-2.915 E
(\214c generator)-.275 E(characterized by a distrib)100.8 652.6 Q
(ution of connection sizes?)-.22 E 2.75(Am)5.5 G(ix of long-li)-2.75 E
-.165(ve)-.275 G 2.75(da).165 G(nd web traf)-2.75 E(\214c,)-.275 E
(with the web traf)100.8 665.6 Q(\214c characterized by a distrib)-.275
E(ution of connection sizes?)-.22 E(Or something)5.5 E(else?)100.8 678.6
Q(Ef)100.8 695.2 Q(fect on congestion control metrics: The distrib)-.275
E(ution of pack)-.22 E(et sequence numbers af)-.11 E(fects)-.275 E
(the b)100.8 708.2 Q(urstiness of aggre)-.22 E -.055(ga)-.165 G(te traf)
.055 E(\214c on a link, thereby af)-.275 E
(fecting all congestion control metrics)-.275 E(for which this is a f)
100.8 721.2 Q(actor)-.11 E 5.5(.A)-.605 G 2.75(sa)-5.5 G 2.75(ne)-2.75 G
(xample, [FK02] illustrates that the traf)-2.915 E(\214c mix can af)
-.275 E(fect)-.275 E(Flo)72 769 Q(yd, K)-.11 E 281.729(ohler Section)
-.385 F 2.75(4. [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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E
(the queue dynamics on a congested link.)100.8 85 Q
([Find more to cite, about the ef)5.5 E(fect of the)-.275 E(distrib)
100.8 98 Q(ution of pack)-.22 E
(et sequence numbers on congestion control metrics.])-.11 E
([Add a paragraph about the impact of medium-size \215o)100.8 114.6 Q
(ws.])-.275 E([Add a paragraph about the impact of \215o)100.8 131.2 Q
(ws starting and stopping.])-.275 E([Add a w)100.8 147.8 Q
(arning about scenarios that use only long-li)-.11 E -.165(ve)-.275 G
2.75<648d>.165 G -.275(ow)-2.75 G(s, or a mix of long-li).275 E -.165
(ve)-.275 G 2.75(da).165 G(nd)-2.75 E -.165(ve)100.8 160.8 S
(ry short \215o).165 E(ws.])-.275 E
(Measurements: [Cite some of the literature.])100.8 177.4 Q -.385(Tr)
100.8 194 S(af).385 E(\214c generators: Some of the a)-.275 E -.275(va)
-.22 G(ilable traf).275 E
(\214c generators are listed on the web site for)-.275 E("T)100.8 207 Q
(raf)-.385 E(\214c Generators for Internet T)-.275 E(raf)-.385 E
(\214c" [TG].)-.275 E(This includes pointers to traf)5.5 E
(\214c generators)-.275 E(for peer)100.8 220 Q(-to-peer traf)-.22 E
(\214c, traf)-.275 E(\214c from online g)-.275 E(ames, and traf)-.055 E
(\214c from Distrib)-.275 E(uted Denial of)-.22 E
(Service \(DDoS\) attacks.)100.8 233 Q(In the NS simulator)100.8 249.6 Q
2.75(,t)-.44 G(he distrib)-2.75 E(ution of pack)-.22 E
(et sequence numbers for TCP pack)-.11 E(ets on a link)-.11 E
(can be reported by the queue monitor at a router)100.8 262.6 Q 5.5(.T)
-.605 G(his is illustrated in the v)-5.5 E(alidation test)-.275 E
("./test-all-simple stats3" in the directory tcl/test.)100.8 275.6 Q/F1
11/Times-Bold@0 SF(5.)72 301.6 Q/F2 14/Times-Bold@0 SF(The Distrib)5.5 E
(ution of P)-.28 E(ack)-.14 E(et Sizes)-.14 E F0
(De\214nition: The distrib)100.8 318.2 Q(ution of pack)-.22 E
(et sizes is de\214ned in a straightforw)-.11 E(ard w)-.11 E(ay)-.11 E
2.75(,u)-.715 G(sing)-2.75 E(pack)100.8 331.2 Q(et sizes in bytes.)-.11
E(Determining f)100.8 347.8 Q(actors: The distrib)-.11 E(ution of pack)
-.22 E(et sizes is determined by the traf)-.11 E(\214c mix, the)-.275 E
(path MTUs, and by the pack)100.8 360.8 Q
(et sizes used by the transport-le)-.11 E -.165(ve)-.275 G 2.75(ls).165
G(enders.)-2.75 E(The distrib)100.8 377.4 Q(ution of pack)-.22 E
(et sizes on a link is also determined by the mix of forw)-.11 E
(ard-path)-.11 E(TCP traf)100.8 390.4 Q(\214c and re)-.275 E -.165(ve)
-.275 G(rse-path TCP traf).165 E
(\214c in that scenario, for a scenario characterized by a)-.275 E
(`forw)100.8 403.4 Q
(ard path' \(e.g., left to right on a particular link\) and a `re)-.11 E
-.165(ve)-.275 G(rse path' \(e.g., right to left).165 E
(on the same link\).)100.8 416.4 Q -.165(Fo)5.5 G 2.75(rs).165 G
(uch a scenario, the forw)-2.75 E(ard-path TCP traf)-.11 E
(\214c contrib)-.275 E(utes data)-.22 E(pack)100.8 429.4 Q
(ets to the forw)-.11 E(ard link and ackno)-.11 E(wledgment pack)-.275 E
(ets to the re)-.11 E -.165(ve)-.275 G(rse link, while the).165 E(re)
100.8 442.4 Q -.165(ve)-.275 G(rse-path TCP traf).165 E(\214c contrib)
-.275 E(utes small ackno)-.22 E(wledgment pack)-.275 E(ets to the forw)
-.11 E(ard link.)-.11 E(The ratio between TCP data and TCP A)100.8 455.4
Q(CK pack)-.44 E(ets on a link can be used as some)-.11 E
(indication of the ratio between forw)100.8 468.4 Q(ard-path and re)-.11
E -.165(ve)-.275 G(rse-path TCP traf).165 E(\214c.)-.275 E(Ef)100.8 485
Q(fect on congestion control metrics: The distrib)-.275 E(ution of pack)
-.22 E(et sizes on a link is an)-.11 E(indicator of the ratio of forw)
100.8 498 Q(ard-path and re)-.11 E -.165(ve)-.275 G(rse-path TCP traf)
.165 E(\214c in that netw)-.275 E 2.75(ork. The)-.11 F(amount of re)
100.8 511 Q -.165(ve)-.275 G(rse-path traf).165 E
(\214c determines the loss and queueing delay e)-.275 E(xperienced by)
-.165 E(ackno)100.8 524 Q(wledgement pack)-.275 E(ets on the re)-.11 E
-.165(ve)-.275 G(rse path, signi\214cantly af).165 E(fecting the b)-.275
E(urstiness of the)-.22 E(aggre)100.8 537 Q -.055(ga)-.165 G(te traf)
.055 E(\214c on the forw)-.275 E(ard path.)-.11 E([In what other w)5.5 E
(ays does the distrib)-.11 E(ution of pack)-.22 E(et)-.11 E(sizes af)
100.8 550 Q(fect congestion control metrics?])-.275 E
(Measurements: There has been a wealth of measurements o)100.8 566.6 Q
-.165(ve)-.165 G 2.75(rt).165 G(ime on the pack)-2.75 E(et size)-.11 E
(distrib)100.8 579.6 Q(ution of traf)-.22 E(\214c [A00], [HMTG01].)-.275
E(These measurements are generally consistent)5.5 E(with a model of rou\
ghly 10% of the TCP connections using an MSS of roughly 500 bytes,)100.8
592.6 Q
(and with the other 90% of TCP connections using an MSS of 1460 bytes.)
100.8 605.6 Q F1(6.)72 631.6 Q F2(The Ratio Between F)5.5 E
(orward-path and Re)-.35 E -.14(ve)-.21 G(rse-path T).14 E(raf\214c)
-1.036 E F0(De\214nition: F)100.8 648.2 Q
(or a scenario characterized by a `forw)-.165 E
(ard path' \(e.g., left to right on a)-.11 E
(particular link\) and a `re)100.8 661.2 Q -.165(ve)-.275 G
(rse path' \(e.g., right to left on the same link\), the ratio between)
.165 E(forw)100.8 674.2 Q(ard-path and re)-.11 E -.165(ve)-.275 G
(rse-path traf).165 E
(\214c can be de\214ned as the ratio between the forw)-.275 E(ard-path)
-.11 E(traf)100.8 687.2 Q(\214c in bps, and the re)-.275 E -.165(ve)
-.275 G(rse-path traf).165 E(\214c in bps.)-.275 E(Determining f)100.8
703.8 Q(actors: The ratio between forw)-.11 E(ard-path and re)-.11 E
-.165(ve)-.275 G(rse-path traf).165 E(\214c is de\214ned)-.275 E(lar)
100.8 716.8 Q(gely by the traf)-.198 E(\214c mix.)-.275 E(Flo)72 769 Q
(yd, K)-.11 E 281.729(ohler Section)-.385 F 2.75(6. [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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E(Ef)100.8 85 Q
(fect on congestion control metrics: Zhang, Shenk)-.275 E
(er and Clark ha)-.11 E .33 -.165(ve s)-.22 H(ho).165 E(wn in 1991 that)
-.275 E(for TCP)100.8 98 Q 2.75(,t)-1.221 G(he amount of re)-2.75 E
-.165(ve)-.275 G(rse-path traf).165 E(\214c af)-.275 E(fects the A)-.275
E(CK compression and pack)-.44 E(et drop)-.11 E(rate for TCP ackno)100.8
111 Q(wledgement pack)-.275 E(ets, signi\214cantly af)-.11 E
(fecting the b)-.275 E(urstiness of TCP traf)-.22 E<8c63>-.275 E
(on the forw)100.8 124 Q(ard path [ZSC91].)-.11 E
(The queueing delay on the re)5.5 E -.165(ve)-.275 G(rse path also af)
.165 E(fects the)-.275 E(performance of delay-based congestion control \
mechanisms, if the delay is computed)100.8 137 Q
(based on round-trip times.)100.8 150 Q(This has been sho)5.5 E
(wn by Grieco and Mascolo in [GM04] and by)-.275 E(Prasad, Jain, and Do)
100.8 163 Q(vrolis in [PJD04].)-.165 E(Measurements: There is a need fo\
r measurements on the range of ratios between forw)100.8 179.6 Q(ard-)
-.11 E(path and re)100.8 192.6 Q -.165(ve)-.275 G(rse-path traf).165 E
(\214c for congested links.)-.275 E(In particular)5.5 E 2.75(,f)-.44 G
(or TCP traf)-2.75 E(\214c tra)-.275 E -.165(ve)-.22 G(rsing).165 E
(congested link X, what is the lik)100.8 205.6 Q(elihood that the ackno)
-.11 E(wledgement traf)-.275 E(\214c will encounter)-.275 E
(congestion \(i.e., queueing delay)100.8 218.6 Q 2.75(,p)-.715 G(ack)
-2.75 E(et drops\) some)-.11 E(where on the re)-.275 E -.165(ve)-.275 G
(rse path as well?).165 E(As discussed in Section 5, the distrib)100.8
235.2 Q(ution of pack)-.22 E(et sizes on a link can be used as an)-.11 E
(indicator of the ratio of forw)100.8 248.2 Q(ard-path and re)-.11 E
-.165(ve)-.275 G(rse-path TCP traf).165 E(\214c in that netw)-.275 E
(ork.)-.11 E/F1 11/Times-Bold@0 SF(7.)72 274.2 Q/F2 14/Times-Bold@0 SF
(The Distrib)5.5 E(ution of P)-.28 E(er)-.28 E(-P)-.518 E(ack)-.14 E
(et P)-.14 E(eak Flo)-.28 E 3.5(wR)-.14 G(ates)-3.5 E F0
(De\214nition: The distrib)100.8 290.8 Q(ution of peak \215o)-.22 E 2.75
(wr)-.275 G(ates is de\214ned by assigning to each pack)-2.75 E(et the)
-.11 E(peak sending rate in bytes per second of that connection, where \
the peak sending rate is)100.8 303.8 Q(de\214ned o)100.8 316.8 Q -.165
(ve)-.165 G 2.75(r0).165 G(.1-second interv)-2.75 E 2.75(als. The)-.275
F(distrib)2.75 E(ution of peak \215o)-.22 E 2.75(wr)-.275 G(ates gi)
-2.75 E -.165(ve)-.275 G 2.75(ss).165 G(ome)-2.75 E
(indication of the ratio of "alpha" and "beta" traf)100.8 329.8 Q
(\214c on a link, where alpha traf)-.275 E(\214c on a)-.275 E
(congested link is de\214ned as traf)100.8 342.8 Q
(\214c with that link at the main bottleneck, while the beta)-.275 E
(traf)100.8 355.8 Q(\214c on the link has a primary bottleneck else)
-.275 E(where along its path [RSB01].)-.275 E(Determining f)100.8 372.4
Q(actors: The distrib)-.11 E(ution of peak \215o)-.22 E 2.75(wr)-.275 G
(ates is determined by \215o)-2.75 E(ws with)-.275 E(bottlenecks else)
100.8 385.4 Q(where along their end-to-end path, e.g., \215o)-.275 E
(ws with lo)-.275 E(w-bandwidth access)-.275 E 2.75(links. The)100.8
398.4 R(distrib)2.75 E(ution of peak \215o)-.22 E 2.75(wr)-.275 G
(ates is also af)-2.75 E(fected by applications with limited)-.275 E
(sending rates.)100.8 411.4 Q(Ef)100.8 428 Q
(fect on congestion control metrics: The distrib)-.275 E
(ution of peak \215o)-.22 E 2.75(wr)-.275 G(ates af)-2.75 E(fects the)
-.275 E -.22(bu)100.8 441 S(rstiness of aggre).22 E -.055(ga)-.165 G
(te traf).055 E(\214c, with lo)-.275 E(w-peak-rate traf)-.275 E
(\214c decreasing the aggre)-.275 E -.055(ga)-.165 G(te).055 E -.22(bu)
100.8 454 S(rstiness, and adding to the traf).22 E(\214c')-.275 E 2.75
(st)-.605 G(ractability)-2.75 E(.)-.715 E(Measurements: [RSB01].)100.8
470.6 Q(The distrib)5.5 E(ution of peak rates can be e)-.22 E
(xpected to change o)-.165 E -.165(ve)-.165 G(r).165 E(time, as there i\
s an increasing number of high-bandwidth access links to the home, and \
of)100.8 483.6 Q(high-bandwidth Ethernet links at w)100.8 496.6 Q
(ork and at other institutions.)-.11 E(Simulators: [F)100.8 513.2 Q(or \
NS, add a pointer to the DelayBox, "http://dirt.cs.unc.edu/delaybox/", \
for)-.165 E(more easily simulating lo)100.8 526.2 Q
(w-bandwidth access links for \215o)-.275 E(ws.])-.275 E -.77(Te)100.8
542.8 S
(stbeds: In testbeds, Dummynet [Dummynet] and NISTNet [NISTNet] pro).77
E(vide)-.165 E(con)100.8 555.8 Q -.165(ve)-.44 G(nient w).165 E
(ays to emulate paths with dif)-.11 E(ferent limited peak rates.)-.275 E
F1(8.)72 581.8 Q F2(The Distrib)5.5 E(ution of T)-.28 E(ransport Pr)
-1.036 E(otocols.)-.252 E F0(De\214nition: The distrib)100.8 598.4 Q
(ution of transport protocols on a congested link is straightforw)-.22 E
(ard,)-.11 E(with each pack)100.8 611.4 Q(et gi)-.11 E -.165(ve)-.275 G
2.75(ni).165 G(ts associated transport protocol \(e.g., TCP)-2.75 E 2.75
(,U)-1.221 G 2.75(DP\). The)-2.75 F(distrib)2.75 E(ution)-.22 E
(is often gi)100.8 624.4 Q -.165(ve)-.275 G 2.75(nb).165 G
(oth in terms of pack)-2.75 E(ets and in terms of bytes.)-.11 E -.165
(Fo)100.8 641 S 2.75(rU).165 G(DP pack)-2.75 E(ets, it might be more he\
lpful to classify them in terms of the port number)-.11 E 2.75(,o)-.44 G
(r)-2.75 E(the assumed application \(e.g., DNS, RIP)100.8 654 Q 2.75(,g)
-1.221 G(ames, W)-2.805 E(indo)-.44 E(ws Media, RealAudio, RealV)-.275 E
(ideo,)-.66 E 2.75(etc.\) [MA)100.8 667 R 2.75(WI]\). Other)-.99 F(traf)
2.75 E(\214c includes ICMP)-.275 E 2.75(,I)-1.221 G(PSEC, and the lik)
-2.75 E 2.75(e. In)-.11 F(the future there could)2.75 E(be traf)100.8
680 Q(\214c from SCTP)-.275 E 2.75(,D)-1.221 G(CCP)-2.75 E 2.75(,o)
-1.221 G 2.75(rf)-2.75 G(rom other transport protocols.)-2.75 E(Ef)100.8
696.6 Q(fect on congestion control metrics: The distrib)-.275 E
(ution of transport protocols af)-.22 E(fects metrics)-.275 E
(relating to the ef)100.8 709.6 Q(fecti)-.275 E -.165(ve)-.275 G
(ness of A).165 E(QM mechanisms on a link.)-.605 E(Flo)72 769 Q(yd, K)
-.11 E 281.729(ohler Section)-.385 F 2.75(8. [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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E
(Measurements: In the past, TCP traf)100.8 85 Q
(\214c has typically consisted of 90% to 95% of the bytes)-.275 E
(on a link [UW02], [U)100.8 98 Q 2.75(A01]. [Get)-.44 F
(updated citations for this.])2.75 E(Measurement studies sho)5.5 E(w)
-.275 E(that TCP traf)100.8 111 Q(\214c from web serv)-.275 E
(ers almost al)-.165 E -.11(wa)-.11 G
(ys uses conformant TCP congestion control).11 E(procedures [MAF05].)
100.8 124 Q/F1 11/Times-Bold@0 SF(9.)72 150 Q/F2 14/Times-Bold@0 SF
(The Synchr)5.5 E(onization Ratio)-.252 E F0
(De\214nition: The synchronization ratio is de\214ned as the de)100.8
166.6 Q(gree of synchronization of loss)-.165 E -2.365 -.275(ev e)100.8
179.6 T(nts between tw).275 E 2.75(oT)-.11 G(CP \215o)-2.75 E
(ws on the same path.)-.275 E
(Thus, the synchronization ratio is de\214ned)5.5 E
(as a characteristic of an end-to-end path.)100.8 192.6 Q
(When one TCP \215o)5.5 E 2.75(wo)-.275 G 2.75(fap)-2.75 G
(air has a loss e)-2.75 E -.165(ve)-.275 G(nt, the).165 E
(synchronization ratio is gi)100.8 205.6 Q -.165(ve)-.275 G 2.75(nb).165
G 2.75(yt)-2.75 G(he fraction of those loss e)-2.75 E -.165(ve)-.275 G
(nts for which the second \215o).165 E(w)-.275 E(has a loss e)100.8
218.6 Q -.165(ve)-.275 G(nt within one round-trip time.).165 E
(Each connection in a \215o)5.5 E 2.75(wp)-.275 G(air has a separate)
-2.75 E(synchronization ratio, and the o)100.8 231.6 Q -.165(ve)-.165 G
(rall synchronization ratio of the pair of \215o).165 E
(ws is the higher)-.275 E(of the tw)100.8 244.6 Q 2.75(or)-.11 G 2.75
(atios. When)-2.75 F
(measuring the synchronization ratio, it is preferable to start the)2.75
E(tw)100.8 257.6 Q 2.75(oT)-.11 G(CP \215o)-2.75 E(ws at slightly dif)
-.275 E(ferent times, with lar)-.275 E(ge recei)-.198 E .33 -.165(ve w)
-.275 H(indo).165 E(ws.)-.275 E(Determining f)100.8 274.2 Q
(actors: The synchronization ratio is determined lar)-.11 E
(gely by the traf)-.198 E(\214c mix on)-.275 E
(the congested link, and by the A)100.8 287.2 Q
(QM mechanism \(or lack of A)-.605 E(QM mechanism\).)-.605 E(Dif)100.8
303.8 Q(ferent types of TCP \215o)-.275 E(ws are also lik)-.275 E
(ely to ha)-.11 E .33 -.165(ve d)-.22 H(if).165 E
(ferent synchronization measures.)-.275 E(E.g., T)100.8 316.8 Q .22 -.11
(wo H)-.88 H(ighSpeed TCP \215o).11 E(ws might ha)-.275 E .33 -.165
(ve h)-.22 H(igher synchronization measures that tw).165 E(o)-.11 E
(Standard TCP \215o)100.8 329.8 Q
(ws on the same path, because of their more aggressi)-.275 E .33 -.165
(ve w)-.275 H(indo).165 E 2.75(wi)-.275 G(ncrease)-2.75 E 2.75
(rates. Raina,)100.8 342.8 R -1.265 -.88(To w)2.75 H(sle).88 E 1.43
-.715(y, a)-.165 H(nd W).715 E(ischik [R)-.44 E(TW05] ha)-.66 E .33
-.165(ve d)-.22 H(iscussed the relationships between).165 E
(synchronization and TCP')100.8 355.8 Q 2.75(si)-.605 G
(ncrease and decrease parameters.)-2.75 E(Ef)100.8 372.4 Q
(fect on congestion control metrics: The synchronization ratio af)-.275
E(fects con)-.275 E -.165(ve)-.44 G -.198(rg).165 G(ence times).198 E
(for high-bandwidth TCPs.)100.8 385.4 Q(Con)5.5 E -.165(ve)-.44 G -.198
(rg).165 G(ence times are kno).198 E(wn to be poor for some high-)-.275
E(bandwidth protocols in en)100.8 398.4 Q(vironments with high le)-.44 E
-.165(ve)-.275 G(ls of synchronization [LS06].).165 E(Ho)100.8 411.4 Q
(we)-.275 E -.165(ve)-.275 G .88 -.44(r, t).165 H
(he scenarios in [LS06] are of a congested link with one-w).44 E
(ay traf)-.11 E(\214c, long-li)-.275 E -.165(ve)-.275 G(d).165 E<8d6f>
100.8 424.4 Q(ws all with the same round-trip time, and Drop-T)-.275 E
(ail queue management at routers.)-.88 E(These are not realistic scenar\
ios; instead, these are the scenarios that I assume w)100.8 437.4 Q
(ould)-.11 E(maximize the de)100.8 450.4 Q
(gree of synchronization between \215o)-.165 E(ws.)-.275 E -.44(Wi)100.8
467 S(schik and McK).44 E(eo)-.275 E(wn [WM05] ha)-.275 E .33 -.165
(ve s)-.22 H(ho).165 E(wn that the le)-.275 E -.165(ve)-.275 G 2.75(lo)
.165 G 2.75(fs)-2.75 G(ynchronization af)-2.75 E(fects the)-.275 E -.22
(bu)100.8 480 S -.275(ff).22 G(er requirements at congested routers.)
.275 E(Baccelli and Hong [BH02] ha)5.5 E .33 -.165(ve a m)-.22 H(odel)
.165 E(sho)100.8 493 Q(wing the ef)-.275 E
(fect of the synchronization ratio on aggre)-.275 E -.055(ga)-.165 G
(te throughput.).055 E(Measurements: Gren)100.8 509.6 Q
(ville Armitage and Qiang Fu ha)-.44 E .33 -.165(ve p)-.22 H
(erformed initial e).165 E(xperiments of)-.165 E
(synchronization in the Internet, using Standard TCP \215o)100.8 522.6 Q
(ws, and ha)-.275 E .33 -.165(ve f)-.22 H(ound v).165 E(ery lo)-.165 E
2.75(wl)-.275 G -2.365 -.275(ev e)-2.75 H(ls).275 E(of synchronization.)
100.8 535.6 Q(In a discussion of the relationship between stability and\
 desynchronization, Raina, T)100.8 552.2 Q -.275(ow)-.88 G(sle).275 E
-.715(y,)-.165 G(and W)100.8 565.2 Q(ischik [R)-.44 E
(TW05] report that "synchronization has been reported ag)-.66 E
(ain and ag)-.055 E(ain in)-.055 E 2.75(simulations". In)100.8 578.2 R
(contrast, synchronization has not been reported ag)2.75 E(ain and ag)
-.055 E(ain in the)-.055 E(real-w)100.8 591.2 Q(orld Internet.)-.11 E
(Appenzeller)100.8 607.8 Q 2.75(,K)-.44 G(eslassy)-3.025 E 2.75(,a)-.715
G(nd McK)-2.75 E(eo)-.275 E(wn in [AKM04] report the follo)-.275 E
(wing: "Flo)-.275 E(ws are not)-.275 E
(synchronized in a backbone router carrying thousands of \215o)100.8
620.8 Q(ws with v)-.275 E(arying R)-.275 E(TTs. Small)-.66 E -.275(va)
100.8 633.8 S(riations in R).275 E(TT or processing time are suf)-.66 E
(\214cient to pre)-.275 E -.165(ve)-.275 G(nt synchronization [QZK01];)
.165 E
(and the absence of synchronization has been demonstrated in real netw)
100.8 646.8 Q(orks [F02,IMD01].")-.11 E
([Appenzeller et al, Sizing Router Buf)100.8 663.4 Q
(fers, reports that synchronization is rare as the)-.275 E
(number of competing \215o)100.8 676.4 Q(ws increases.)-.275 E -.275
(Kev)5.5 G(in Jef).275 E -.11(fa)-.275 G 2.75(yh).11 G
(as some results on synchronization)-2.75 E(also.])100.8 689.4 Q
(Needed: W)100.8 706 Q 2.75(en)-.88 G
(eed measurements of the synchronization ratio for \215o)-2.75 E
(ws that use high-)-.275 E(bandwidth protocols o)100.8 719 Q -.165(ve)
-.165 G 2.75(rh).165 G(igh-bandwidth paths, gi)-2.75 E -.165(ve)-.275 G
2.75(nt).165 G(ypical le)-2.75 E -.165(ve)-.275 G(ls of competing traf)
.165 E<8c63>-.275 E(Flo)72 769 Q(yd, K)-.11 E 281.729(ohler Section)
-.385 F 2.75(9. [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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E
(and with typical queueing mechanisms at routers \(whate)100.8 85 Q
-.165(ve)-.275 G 2.75(rt).165 G(hese are\), to see if there are)-2.75 E
(higher le)100.8 98 Q -.165(ve)-.275 G(ls of synchronization with high-\
bandwidth protocols such as HighSpeed TCP).165 E(,)-1.221 E -.165(Fa)
100.8 111 S(st TCP).165 E 2.75(,a)-1.221 G(nd the lik)-2.75 E
(e, which are more aggressi)-.11 E .33 -.165(ve t)-.275 H
(han Standard TCP).165 E 5.5(.T)-1.221 G(he assumption)-5.5 E -.11(wo)
100.8 124 S(uld be that in man).11 E 2.75(ye)-.165 G -.44(nv)-2.75 G
(ironments, high-bandwidth protocols ha).44 E .33 -.165(ve h)-.22 H
(igher le).165 E -.165(ve)-.275 G(ls of).165 E
(synchronization than \215o)100.8 137 Q(ws using Standard TCP)-.275 E(.)
-1.221 E/F1 11/Times-Bold@0 SF(10.)72 163 Q/F2 14/Times-Bold@0 SF(Dr)5.5
E(op or Mark Rates as a Function of P)-.252 E(ack)-.14 E(et Size)-.14 E
F0(De\214nition: Drop rates as a function of pack)100.8 179.6 Q
(et size are de\214ned by the actual drop rates for)-.11 E(dif)100.8
192.6 Q(ferent pack)-.275 E
(ets on an end-to-end path or on a congested link o)-.11 E -.165(ve)
-.165 G 2.75(rap).165 G(articular time)-2.75 E(interv)100.8 205.6 Q 2.75
(al. In)-.275 F(some cases, e.g., Drop-T)2.75 E
(ail queues in units of pack)-.88 E(ets, general statements can)-.11 E
(be made; e.g., that lar)100.8 218.6 Q(ge and small pack)-.198 E
(ets will e)-.11 E(xperience the same pack)-.165 E(et drop rates.)-.11 E
(Ho)100.8 231.6 Q(we)-.275 E -.165(ve)-.275 G .88 -.44(r, i).165 H 2.75
(no).44 G(ther cases, e.g., Drop-T)-2.75 E
(ail queues in units of bytes, no such general statement)-.88 E
(can be made, and the drop rate as a function of pack)100.8 244.6 Q
(et size will be determined in part by the)-.11 E(traf)100.8 257.6 Q
(\214c mix at the congested link at that point of time.)-.275 E
(Determining f)100.8 274.2 Q
(actors: The drop rate as a function of pack)-.11 E
(et size is determined in part by the)-.11 E(queue architecture.)100.8
287.2 Q(E.g., is the Drop-T)5.5 E(ail queue in units of pack)-.88 E
(ets, of bytes, of 60-byte)-.11 E -.22(bu)100.8 300.2 S -.275(ff).22 G
(ers, or of a mix of b).275 E(uf)-.22 E(fer sizes?)-.275 E(Is the A)5.5
E(QM mechanism in pack)-.605 E(et mode, dropping each)-.11 E(pack)100.8
313.2 Q(et with the same probability)-.11 E 2.75(,o)-.715 G 2.75(ri)
-2.75 G 2.75(nb)-2.75 G(yte mode, with the probability of dropping or)
-2.75 E(marking a pack)100.8 326.2 Q(et being proportional to the pack)
-.11 E(et size in bytes.)-.11 E(The ef)100.8 342.8 Q(fect of pack)-.275
E(et size on drop rate w)-.11 E(ould also be af)-.11 E
(fected by the presence of preferential)-.275 E
(scheduling for small pack)100.8 355.8 Q(ets, or by dif)-.11 E
(ferential scheduling for pack)-.275 E(ets from dif)-.11 E(ferent \215o)
-.275 E(ws)-.275 E(\(e.g., per)100.8 368.8 Q(-\215o)-.22 E 2.75(ws)-.275
G(cheduling, or dif)-2.75 E(ferential scheduling for UDP and TCP traf)
-.275 E(\214c\).)-.275 E(In man)100.8 385.4 Q 2.75(ye)-.165 G -.44(nv)
-2.75 G(ironments, the drop rate as a function of pack).44 E
(et size will be hea)-.11 E(vily af)-.22 E(fected by)-.275 E(the traf)
100.8 398.4 Q(\214c mix at a particular time.)-.275 E -.165(Fo)5.5 G
2.75(re).165 G(xample, is the traf)-2.915 E(\214c mix dominated by lar)
-.275 E(ge)-.198 E(pack)100.8 411.4 Q(ets, or by smaller ones?)-.11 E
(In some cases, the o)5.5 E -.165(ve)-.165 G(rall pack).165 E
(et drop rate could also af)-.11 E(fect)-.275 E(the relati)100.8 424.4 Q
.33 -.165(ve d)-.275 H(rop rates for dif).165 E(ferent pack)-.275 E
(et sizes.)-.11 E(In wireless netw)100.8 441 Q
(orks, the drop rate as a function of pack)-.11 E
(et size is also determined by the)-.11 E(pack)100.8 454 Q
(et corruption rate as a function of pack)-.11 E(et size.)-.11 E
([Cite Deborah Pinck')5.5 E 2.75(sp)-.605 G(apers on)-2.75 E(Satellite-\
Enhanced Personal Communications Experiments and on Experimental Result\
s)100.8 467 Q(from Internetw)100.8 480 Q(orking Data Applications Ov)
-.11 E(er V)-.165 E(arious W)-1.221 E(ireless Netw)-.44 E
(orks Using a Single)-.11 E(Fle)100.8 493 Q
(xible Error Control Protocol.])-.165 E([Cite the general literature.])
5.5 E(Ef)100.8 509.6 Q(fect on congestion control metrics: The drop rat\
e as a function of pack)-.275 E(et size has a)-.11 E(signi\214cant ef)
100.8 522.6 Q(fect on the performance of congestion control for V)-.275
E(oIP and other small-pack)-1.419 E(et)-.11 E<8d6f>100.8 535.6 Q 2.75
(ws. [Citation:)-.275 F("TFRC for V)2.75 E(oice: the V)-1.419 E(oIP V)
-1.419 E(ariant", draft-ietf-dccp-tfrc-v)-1.221 E(oip-02.txt, and)-.22 E
(earlier papers.])100.8 548.6 Q(The drop rate as a function of pack)5.5
E(et size also has an ef)-.11 E(fect on TCP)-.275 E
(performance, as it af)100.8 561.6 Q(fects the drop rates for TCP')-.275
E 2.75(sS)-.605 G(YN and A)-2.75 E(CK pack)-.44 E 2.75(ets. [Citation:)
-.11 F(Jef)100.8 574.6 Q -.11(fa)-.275 G 2.75(ya).11 G(nd others.])-2.75
E(Measurements: W)100.8 591.2 Q 2.75(en)-.88 G
(eed measurements of the drop rate as a function of pack)-2.75 E
(et size o)-.11 E -.165(ve)-.165 G 2.75(ra).165 G
(wide range of paths, or for a wide range of congested links.)100.8
604.2 Q -.165(Fo)5.5 G 2.75(rt).165 G(ests of relati)-2.75 E .33 -.165
(ve d)-.275 H(rop rates).165 E(on end-to-end pack)100.8 617.2 Q
(ets, one possibility w)-.11 E(ould be to run successi)-.11 E .33 -.165
(ve T)-.275 H(CP connections with).165 E
(200-byte, 512-byte, and 1460-byte pack)100.8 630.2 Q
(ets, and to compare the pack)-.11 E(et drop rates.)-.11 E(The)5.5 E
(ideal test w)100.8 643.2 Q
(ould include running TCP connections on the re)-.11 E -.165(ve)-.275 G
(rse path, to measure the drop).165 E(rates for the small A)100.8 656.2
Q(CK pack)-.44 E(ets on the forw)-.11 E(ard path.)-.11 E(It w)5.5 E
(ould also be useful to)-.11 E(characterize the dif)100.8 669.2 Q
(ference in drop rates for 200-byte TCP pack)-.275 E
(ets and 200-byte UDP)-.11 E(pack)100.8 682.2 Q(ets, e)-.11 E -.165(ve)
-.275 G 2.75(nt).165 G(hough some of this dif)-2.75 E
(ference could be due to the relati)-.275 E .33 -.165(ve b)-.275 H
(urstiness of the)-.055 E(dif)100.8 695.2 Q(ferent connections.)-.275 E
(Ping e)100.8 711.8 Q(xperiments could also be used to get measurements\
 of drop rates as a function size,)-.165 E -.22(bu)100.8 724.8 S 2.75
(ti).22 G 2.75(tw)-2.75 G(ould be necessary to mak)-2.86 E 2.75(es)-.11
G(ure that the ping sending rates were adjusted to be TCP-)-2.75 E(Flo)
72 769 Q(yd, K)-.11 E 276.229(ohler Section)-.385 F 2.75(10. [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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E(friendly)100.8 85 Q(.)-.715 E
([Cite the kno)100.8 101.6 Q
(wn literature on drop rates as a function of pack)-.275 E(et size.])
-.11 E(Our conjecture is that there is a wide range of beha)100.8 118.2
Q(viors for this characteristic in the real)-.22 E -.11(wo)100.8 131.2 S
2.75(rld. Routers).11 F(include Drop-T)2.75 E(ail queues in pack)-.88 E
(ets, bytes, and b)-.11 E(uf)-.22 E(fer sizes in between;)-.275 E
(these will ha)100.8 144.2 Q .33 -.165(ve q)-.22 H(uite dif).165 E
(ferent drop rates as a function of pack)-.275 E(et size.)-.11 E
(Some routers include)5.5 E(RED in byte mode \(the def)100.8 157.2 Q
(ault for RED in Linux\) and some ha)-.11 E .33 -.165(ve R)-.22 H
(ED in pack).165 E(et mode)-.11 E(\(Cisco, I belie)100.8 170.2 Q -.165
(ve)-.275 G 2.75(\). This).165 F(also af)2.75 E
(fects drop rates as a function of pack)-.275 E(et size.)-.11 E
(Some routers on congested access links use per)100.8 186.8 Q(-\215o)
-.22 E 2.75(ws)-.275 G 2.75(cheduling. In)-2.75 F
(this case, does the per)2.75 E(-)-.22 E<8d6f>100.8 199.8 Q 2.75(ws)
-.275 G(cheduling ha)-2.75 E .33 -.165(ve t)-.22 H(he goal of f).165 E
(airness in *bytes* per second or in *pack)-.11 E(ets* per second?)-.11
E(What ef)100.8 212.8 Q(fect does the per)-.275 E(-\215o)-.22 E 2.75(ws)
-.275 G(cheduling ha)-2.75 E .33 -.165(ve o)-.22 H 2.75(nt).165 G
(he drop rate as a function of pack)-2.75 E(et size,)-.11 E(for pack)
100.8 225.8 Q(ets in dif)-.11 E(ferent \215o)-.275 E
(ws \(e.g., a small-pack)-.275 E(et V)-.11 E(oIP \215o)-1.419 E 2.75(wc)
-.275 G(ompeting ag)-2.75 E(ainst a lar)-.055 E(ge-)-.198 E(pack)100.8
238.8 Q(et TCP \215o)-.11 E(w\) or for pack)-.275 E
(ets within the same \215o)-.11 E 2.75(w\()-.275 G(small A)-2.75 E
(CK pack)-.44 E(ets and lar)-.11 E(ge data)-.198 E(pack)100.8 251.8 Q
(ets on a tw)-.11 E(o-w)-.11 E(ay TCP connection\).)-.11 E/F1 11
/Times-Bold@0 SF(11.)72 277.8 Q/F2 14/Times-Bold@0 SF(Dr)5.5 E
(op Rates as a Function of Burst Size.)-.252 E F0
(De\214nition: Burst-tolerance, or drop rates as a function of b)100.8
294.4 Q(urst size, can be de\214ned in terms)-.22 E
(of an end-to-end path, or in terms of aggre)100.8 307.4 Q -.055(ga)
-.165 G(te traf).055 E(\214c on a congested link.)-.275 E(The b)100.8
324 Q(urst-tolerance of an end-to-end path is de\214ned in terms of con\
nections with dif)-.22 E(ferent)-.275 E(de)100.8 337 Q(grees of b)-.165
E(urstiness within a round-trip time.)-.22 E(When pack)5.5 E
(ets are sent in b)-.11 E(ursts of N)-.22 E(pack)100.8 350 Q
(ets, does the drop rate v)-.11 E(ary as a function of N?)-.275 E -.165
(Fo)5.5 G 2.75(re).165 G(xample, if the TCP sender sends)-2.915 E
(small b)100.8 363 Q(ursts of K pack)-.22 E
(ets, for K less than the congestion windo)-.11 E 1.43 -.715(w, h)-.275
H .55 -.275(ow d).715 H(oes the size of K).275 E(af)100.8 376 Q
(fect the loss rate?)-.275 E(Similarly)5.5 E 2.75(,f)-.715 G
(or a ping tool sending pings at a certain rate in pack)-2.75 E(ets per)
-.11 E(second, one could see ho)100.8 389 Q 2.75(wt)-.275 G
(he clustering of the ping pack)-2.75 E(ets in clusters of size K af)
-.11 E(fects)-.275 E(the pack)100.8 402 Q(et drop rate.)-.11 E(As al)5.5
E -.11(wa)-.11 G(ys with such ping e).11 E(xperiments, it w)-.165 E
(ould be important to)-.11 E
(adjust the sending rate to maintain a longer)100.8 415 Q
(-term sending rate that w)-.22 E(as TCP-friendly)-.11 E(.)-.715 E
(Determining f)100.8 431.6 Q(actors: The b)-.11 E
(urst-tolerance is determined lar)-.22 E(gely by the A)-.198 E
(QM mechanisms)-.605 E
(for the congested routers on a path, and by the traf)100.8 444.6 Q
(\214c mix.)-.275 E -.165(Fo)5.5 G 2.75(raD).165 G(rop-T)-2.75 E
(ail queue with only)-.88 E 2.75(as)100.8 457.6 S
(mall number of competing \215o)-2.75 E(ws, the b)-.275 E
(urst-tolerance is lik)-.22 E(ely to be lo)-.11 E 1.43 -.715(w, a)-.275
H(nd for A).715 E(QM)-.605 E(mechanisms where the pack)100.8 470.6 Q
(et drop rate is a function of the a)-.11 E -.165(ve)-.22 G
(rage queue size rather than).165 E(the instantaneous queue size, the b)
100.8 483.6 Q(urst tolerance should be quite high.)-.22 E(Ef)100.8 500.2
Q(fect on congestion control metrics: The b)-.275 E
(urst-tolerance of the path or congested link can)-.22 E(af)100.8 513.2
Q(fect f)-.275 E(airness between competing \215o)-.11 E(ws with dif)
-.275 E(ferent round-trip times; for e)-.275 E(xample,)-.165 E
(Standard TCP \215o)100.8 526.2 Q
(ws with longer round-trip times are lik)-.275 E(ely to ha)-.11 E .33
-.165(ve a m)-.22 H(ore b).165 E(ursty arri)-.22 E -.275(va)-.275 G(l)
.275 E(pattern at the congested link that that of Standard TCP \215o)
100.8 539.2 Q(ws with shorter round-trip times.)-.275 E
(As a result, in en)100.8 552.2 Q(vironment with lo)-.44 E 2.75(wb)-.275
G(urst tolerance \(e.g., scenarios with Drop-T)-2.97 E(ail)-.88 E
(queues\), longer)100.8 565.2 Q
(-round-trip-time TCP connections can see higher pack)-.22 E
(et drop rates than)-.11 E(other TCP connections, and recei)100.8 578.2
Q .33 -.165(ve a)-.275 H 2.75(ne).165 G -.165(ve)-3.025 G 2.75(ns).165 G
(maller fraction of the link bandwidth than)-2.75 E(the)100.8 591.2 Q
2.75(yw)-.165 G(ould otherwise.)-2.86 E([FJ92] \(Section 3.2\).)5.5 E
1.76 -.88(We n)5.5 H(ote that some TCP traf).88 E(\214c is inherently)
-.275 E -.22(bu)100.8 604.2 S(rsty).22 E 2.75(,e)-.715 G(.g., Standard \
TCP without rate-based pacing, particularly in the presence of)-2.75 E
(dropped A)100.8 617.2 Q(CK pack)-.44 E(ets or of A)-.11 E
(CK compression.)-.44 E(The b)5.5 E(urst-tolerance of a router can also)
-.22 E(af)100.8 630.2 Q(fect the delay-throughput tradeof)-.275 E
(fs and pack)-.275 E(et drop rates of the path or of the congested)-.11
E(link.)100.8 643.2 Q(Measurements: One could measure the b)100.8 659.8
Q(urst-tolerance of an end-to-end path by running)-.22 E(successi)100.8
672.8 Q .33 -.165(ve T)-.275 H(CP connections, forcing b).165 E
(ursts of size at least K by dropping an appropriate)-.22 E
(fraction of the A)100.8 685.8 Q(CK pack)-.44 E(ets to the TCP recei)
-.11 E -.165(ve)-.275 G 3.96 -.605(r. A).165 H(lternately).605 E 2.75
(,i)-.715 G 2.75(fo)-2.75 G(ne had control of the TCP)-2.75 E(sender)
100.8 698.8 Q 2.75(,o)-.44 G(ne could modify the TCP sender to send b)
-2.75 E(ursts of K pack)-.22 E(ets when the congestion)-.11 E(windo)
100.8 711.8 Q 2.75(ww)-.275 G(as K or more se)-2.86 E(gments.)-.165 E
(Flo)72 769 Q(yd, K)-.11 E 270.729(ohler Section)-.385 F 2.75(11. [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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E(Blanton and Allman in [B)100.8
85 Q(A05] consider the TCP micro-b)-.385 E
(ursts that result from the receipt)-.22 E(of a single ackno)100.8 98 Q
(wledgement pack)-.275 E
(et or from application-layer dynamics, and consider)-.11 E -.22(bu)
100.8 111 S(rsts of four or more pack).22 E 2.75(ets. The)-.11 F 2.75
(yc)-.165 G(onsider four traces, and plot the probability of at)-2.75 E
(least one pack)100.8 124 Q(et from a b)-.11 E
(urst being lost, as a function of b)-.22 E(urst size.)-.22 E
(Considering only)5.5 E(connections with both b)100.8 137 Q
(ursts and pack)-.22 E(et losses, the probability of pack)-.11 E
(et loss when the)-.11 E(TCP connection w)100.8 150 Q(as b)-.11 E
(ursting w)-.22 E(as some)-.11 E
(what higher than the probability of pack)-.275 E(et loss)-.11 E
(when the TCP connection w)100.8 163 Q(as not b)-.11 E
(ursting in three of the four traces.)-.22 E -.165(Fo)5.5 G 2.75(re).165
G(ach trace, the)-2.75 E(paper sho)100.8 176 Q(ws the aggre)-.275 E
-.055(ga)-.165 G(te probability of loss as a function of the b).055 E
(urst size in pack)-.22 E(ets.)-.11 E(Because these are aggre)100.8 189
Q -.055(ga)-.165 G
(te statistics, it cannot be determined if there is a correlation).055 E
(between the b)100.8 202 Q(urst size and the TCP connection')-.22 E 2.75
(ss)-.605 G(ending rate.)-2.75 E
([Look at: M. Allman and E. Blanton, "Notes on Burst Mitig)100.8 218.6 Q
(ation for T)-.055 E(ransport Protocols",)-.385 E -.44(AC)100.8 231.6 S
2.75(MC).44 G(omputer Communication Re)-2.75 E(vie)-.275 E 1.43 -.715
(w, v)-.275 H(ol. 35\(2\), \(2005\).]).495 E
(Making inferences about the A)100.8 248.2 Q
(QM mechanism for the congested router on an end-to-end)-.605 E
(path: One potential use of measurement tools for determining the b)
100.8 261.2 Q(urst-tolerance of an end-)-.22 E(to-end path w)100.8 274.2
Q(ould be to mak)-.11 E 2.75(ei)-.11 G
(nferences about the presence or absence of an A)-2.75 E(QM)-.605 E
(mechanism at the congested link or links.)100.8 287.2 Q
(As a simple test, one could run a TCP)5.5 E
(connection until the connection comes out of slo)100.8 300.2 Q 2.75
(w-start. If)-.275 F(the recei)2.75 E .33 -.165(ve w)-.275 H(indo).165 E
2.75(wo)-.275 G 2.75(ft)-2.75 G(he TCP)-2.75 E(connection w)100.8 313.2
Q(as suf)-.11 E(\214ciently high that the connection e)-.275 E
(xited slo)-.165 E(w-start with pack)-.275 E(et drops or)-.11 E
(marks instead of because of the limitation of the recei)100.8 326.2 Q
.33 -.165(ve w)-.275 H(indo).165 E 1.43 -.715(w, o)-.275 H
(ne could record the).715 E(congestion windo)100.8 339.2 Q 2.75(wa)-.275
G 2.75(tt)-2.75 G(he end of slo)-2.75 E(w-start, and the number of pack)
-.275 E(ets dropped from this)-.11 E(windo)100.8 352.2 Q 4.18 -.715
(w. A)-.275 H(high pack)3.465 E
(et drop rate might be more typical of a Drop-T)-.11 E
(ail queue with small-)-.88 E(scale statistical multiple)100.8 365.2 Q
(xing on the congested link, and a single pack)-.165 E
(et drop coming out of)-.11 E(slo)100.8 378.2 Q(w-start w)-.275 E
(ould suggest an A)-.11 E(QM mechanism at the congested link.)-.605 E
(The synchronization measure could also add information about the lik)
100.8 394.8 Q(ely presence or)-.11 E(absence of A)100.8 407.8 Q(QM on t\
he congested link\(s\) of an end-to-end path, with paths with higher)
-.605 E(le)100.8 420.8 Q -.165(ve)-.275 G
(ls of synchronization being more lik).165 E(ely to ha)-.11 E .33 -.165
(ve D)-.22 H(rop-T).165 E(ail queues with small-scale)-.88 E
(statistical multiple)100.8 433.8 Q(xing on the congested link\(s\).)
-.165 E(Lui and Cro)100.8 450.4 Q -.165(ve)-.165 G
(lla in [LC01] use loss pairs to infer the queue size when pack).165 E
(ets are dropped.)-.11 E 2.75(Al)100.8 463.4 S(oss pair consists of tw)
-2.75 E 2.75(op)-.11 G(ack)-2.75 E
(ets sent back-to-back, where one of the tw)-.11 E 2.75(op)-.11 G(ack)
-2.75 E(ets is)-.11 E(dropped in the netw)100.8 476.4 Q 2.75(ork. The)
-.11 F(round-trip time of the survi)2.75 E(ving pack)-.275 E
(et is used to estimate the)-.11 E
(round-trip time when the companion pack)100.8 489.4 Q(et w)-.11 E
(as dropped in the netw)-.11 E 2.75(ork. F)-.11 F(or a path with)-.165 E
(Drop-T)100.8 502.4 Q(ail queueing at the congested link, this round-tr\
ip time can be used to estimate the)-.88 E(queue size, gi)100.8 515.4 Q
-.165(ve)-.275 G 2.75(ne).165 G
(stimates of the link bandwidth and minimum round-trip time.)-2.75 E
-.165(Fo)5.5 G 2.75(rap).165 G(ath)-2.75 E(with A)100.8 528.4 Q(QM at t\
he congested link, trial pairs are also considered, where a trial pair \
is an)-.605 E(y)-.165 E(pair of pack)100.8 541.4 Q
(ets sent back-to-back.)-.11 E
([LC01] uses the ratio between the number of loss pairs)5.5 E(and the n\
umber of trial pairs for each round-trip range to estimate the drop pro\
bability of)100.8 554.4 Q(the A)100.8 567.4 Q
(QM mechanism at the congested link as a function of queue size.)-.605 E
([LC01] uses loss)5.5 E(pairs in simulation settings with a minimum of \
noise in terms of queueing delays else)100.8 580.4 Q(where)-.275 E
(on the forw)100.8 593.4 Q(ard or re)-.11 E -.165(ve)-.275 G(rse path.)
.165 E([Cite the rele)100.8 610 Q -.275(va)-.275 G
(nt literature about tools for determining the A).275 E
(QM mechanism on an end-to-)-.605 E(end path.])100.8 623 Q/F1 11
/Times-Bold@0 SF(12.)72 649 Q/F2 14/Times-Bold@0 SF(Dr)5.5 E
(op Rates as a Function of Sending Rate.)-.252 E F0(De\214nition: Drop \
rates as a function of sending rate is de\214ned in terms of the drop b\
eha)100.8 665.6 Q(vior)-.22 E(of a \215o)100.8 678.6 Q 2.75(wi)-.275 G
2.75(nt)-2.75 G(he end-to-end path.)-2.75 E
(That is, does the sending rate of an indi)5.5 E(vidual \215o)-.275 E
2.75(wa)-.275 G -.275(ff)-2.75 G(ect).275 E(its o)100.8 691.6 Q(wn pack)
-.275 E(et drop rate, or its pack)-.11 E(et drop rate lar)-.11 E
(gely independent of the sending rate of)-.198 E(the \215o)100.8 704.6 Q
(w?)-.275 E(Flo)72 769 Q(yd, K)-.11 E 270.729(ohler Section)-.385 F 2.75
(12. [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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E(Determining f)100.8 85 Q
(actors: The sending rate of the \215o)-.11 E 2.75(wa)-.275 G -.275(ff)
-2.75 G(ects its o).275 E(wn pack)-.275 E(et drop rate in an)-.11 E(en)
100.8 98 Q(vironment with small-scale statistical multiple)-.44 E
(xing on the congested link.)-.165 E(The pack)5.5 E(et)-.11 E
(drop rate is lar)100.8 111 Q
(gely independent of the sending rate in an en)-.198 E
(vironment with lar)-.44 E(ge-scale)-.198 E(statistical multiple)100.8
124 Q(xing, with man)-.165 E 2.75(yc)-.165 G(ompeting small \215o)-2.75
E(ws at the congested link.)-.275 E(Thus, the)5.5 E(beha)100.8 137 Q(vi\
or of drop rates as a function of sending rate is a rough measure of th\
e le)-.22 E -.165(ve)-.275 G 2.75(lo).165 G(f)-2.75 E
(statistical multiple)100.8 150 Q
(xing on the congested links of an end-to-end path.)-.165 E(Ef)100.8
166.6 Q(fect on congestion control metrics: The le)-.275 E -.165(ve)
-.275 G 2.75(lo).165 G 2.75(fs)-2.75 G(tatistical multiple)-2.75 E
(xing at the congested)-.165 E(link can af)100.8 179.6 Q(fect the perfo\
rmance of congestion control mechanisms in transport protocols.)-.275 E
-.165(Fo)100.8 192.6 S 2.75(re).165 G
(xample, delay-based congestion control is often better suited to en)
-2.915 E(vironments small-)-.44 E(scale statistical multiple)100.8 205.6
Q(xing at the congested link, where the transport protocol responds to)
-.165 E(the delay caused by its o)100.8 218.6 Q(wn sending rate.)-.275 E
(Measurements: In a simulation or testbed, the le)100.8 235.2 Q -.165
(ve)-.275 G 2.75(lo).165 G 2.75(fs)-2.75 G(tatistical multiple)-2.75 E
(xing on the)-.165 E(congested link can be observ)100.8 248.2 Q
(ed directly)-.165 E 5.5(.I)-.715 G 2.75(nt)-5.5 G(he Internet, the le)
-2.75 E -.165(ve)-.275 G 2.75(lo).165 G 2.75(fs)-2.75 G
(tatistical multiple)-2.75 E(xing)-.165 E(on the congested links of an \
end-to-end path can be inferred indirectly through per)100.8 261.2 Q
(-\215o)-.22 E(w)-.275 E(measurements, by observing whether the pack)
100.8 274.2 Q(et drop rate v)-.11 E(aries as a function of the)-.275 E
(sending rate of the \215o)100.8 287.2 Q -.715(w.)-.275 G/F1 11
/Times-Bold@0 SF(13.)72 313.2 Q/F2 14/Times-Bold@0 SF(Congestion Contr)
5.5 E(ol Mechanisms f)-.252 E(or T)-.35 E
(raf\214c, along with Sender and)-1.036 E(Recei)72 326.2 Q -.14(ve)-.14
G 3.5(rB).14 G(uffer Sizes.)-3.5 E F0(Ef)100.8 342.8 Q
(fect on congestion control metrics: Please don')-.275 E 2.75(te)-.198 G
-.275(va)-3.025 G(luate A).275 E(QM mechanisms by using)-.605 E
(Reno TCP)100.8 355.8 Q 2.75(,o)-1.221 G 2.75(re)-2.75 G -.275(va)-3.025
G(luate ne).275 E 2.75(wt)-.275 G
(ransport protocols by comparing them with the performance of)-2.75 E
(Reno TCP!)100.8 368.8 Q -.165(Fo)5.5 G 2.75(ra).165 G 2.75(ne)-2.75 G
(xplanation, see [FK02] \(Section 3.4\).)-2.915 E 2.75
(Measurements: See)100.8 385.4 R([MAF05].)2.75 E F1(14.)72 411.4 Q F2
(Characterization of Congested Links in T)5.5 E(erms of Band)-1.288 E
(width and)-.21 E -1.036(Ty)72 424.4 S(pical Le)1.036 E -.14(ve)-.21 G
(ls of Congestion).14 E F1(14.1.)72 463.4 Q/F3 13/Times-Bold@0 SF(Band)
5.5 E(width)-.195 E F1(14.2.)72 502.4 Q F3(Queue Management Mechanisms)
5.5 E F1(14.3.)72 541.4 Q F3 -.962(Ty)5.5 G(pical Le).962 E -.13(ve)
-.195 G(ls of Congestion).13 E F0
([Pointers to the current state of our kno)100.8 558 Q(wledge.])-.275 E
F1(15.)72 584 Q F2(Characterization of Challenging Lo)5.5 E(wer Lay)-.14
E(ers.)-.14 E F0 -.44(Wi)100.8 600.6 S
(th an increasing number of wireless netw).44 E
(orks connecting to the wired Internet, more and)-.11 E(more end-to-end\
 paths will contain a combination of wired and wireless links.)100.8
613.6 Q(These)5.5 E(wireless links e)100.8 626.6 Q(xhibit ne)-.165 E
2.75(wc)-.275 G
(haracteristics which congestion control mechanisms will need)-2.75 E
(to cope with.)100.8 639.6 Q(The main characteristics, detailed in subs\
equent sections, include error)5.5 E(losses, pack)100.8 652.6 Q
(et reordering, delay v)-.11 E(ariation, bandwidth v)-.275 E
(ariation, and bandwidth and latenc)-.275 E(y)-.165 E(asymmetry)100.8
665.6 Q(.)-.715 E F1(15.1.)72 691.6 Q F3(Err)5.5 E(or Losses)-.234 E F0
2.75(De\214nition: P)100.8 717.6 R(ack)-.165 E
(et losses due to corruption rarely occur on wired links, b)-.11 E
(ut occur on)-.22 E(Flo)72 769 Q(yd, K)-.11 E 262.479(ohler Section)
-.385 F 2.75(15.1. [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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E
(wireless links due to random/transient errors and/or e)100.8 85 Q
(xtended b)-.165 E(urst errors.)-.22 E(If pack)5.5 E(et errors)-.11 E
(cannot be detected and discarded within the netw)100.8 98 Q
(ork through error detection schemes or)-.11 E(reco)100.8 111 Q -.165
(ve)-.165 G(red through error reco).165 E -.165(ve)-.165 G
(ry schemes such as F).165 E(orw)-.165 E
(ard Error Correction \(FEC\) and)-.11 E
(Automatic Repeat Request \(ARQ\), the corrupted pack)100.8 124 Q
(et is discarded, resulting in an error)-.11 E(loss.)100.8 137 Q
(Determining F)100.8 163 Q 2.75(actors: Error)-.165 F
(losses are primarily caused by the de)2.75 E
(gradation of the quality of)-.165 E 2.75(aw)100.8 176 S
(ireless link \(multipath, f)-2.75 E(ade, etc.\).)-.11 E
(Link errors can be characterized by the type of errors)5.5 E
(that occur \(e.g., random, b)100.8 189 Q
(urst\), the length of time the)-.22 E 2.75(yo)-.165 G(ccur)-2.75 E 2.75
(,a)-.44 G(nd the frequenc)-2.75 E 2.75(ya)-.165 G 2.75(tw)-2.75 G(hich)
-2.75 E(the)100.8 202 Q 2.75(yo)-.165 G(ccur)-2.75 E 5.5(.T)-.605 G(hes\
e characteristics are highly dependent on the wireless channel conditio\
ns)-5.5 E(and are in\215uenced by the distance between tw)100.8 215 Q
2.75(on)-.11 G(odes on a wireless link, the type and)-2.75 E
(orientation of antennas, encoding algorithms, and other f)100.8 228 Q
(actors [22].)-.11 E(Therefore, error)5.5 E
(losses are signi\214cantly in\215uenced by these link errors.)100.8 241
Q(Ef)100.8 267 Q(fect on congestion control metrics:)-.275 E
(Since error losses can be unrelated to congestion,)5.5 E
(congestion control mechanisms should reco)100.8 280 Q -.165(ve)-.165 G
2.75(rf).165 G(rom these types of losses dif)-2.75 E(ferently than)-.275
E(from congestion losses.)100.8 293 Q
(If congestion control mechanisms misinterpret error losses as)5.5 E
(congestion losses, then the)100.8 306 Q 2.75(yr)-.165 G
(espond inappropriately)-2.75 E 2.75(,r)-.715 G
(educing the sending rate too much)-2.75 E 2.75([23]. As)100.8 319 R
2.75(ar)2.75 G
(esult, an unnecessary reduction in the sending rate can occur)-2.75 E
2.75(,w)-.44 G(hen in reality)-2.75 E(the a)100.8 332 Q -.275(va)-.22 G
(ilable bandwidth has not changed.).275 E
(This can result in a reduction in throughput and)5.5 E
(underutilization of the channel.)100.8 345 Q(Ho)5.5 E(we)-.275 E -.165
(ve)-.275 G .88 -.44(r, e).165 H(rror reco).44 E -.165(ve)-.165 G
(ry mechanisms such as FEC or ARQ).165 E(are hea)100.8 358 Q
(vily used in cellular netw)-.22 E
(orks to reduce the impact of error losses [IMLGK03].)-.11 E 2.75
(Measurements: In)100.8 384 R(3G cellular netw)2.75 E(orks, error reco)
-.11 E -.165(ve)-.165 G(ry mechanisms ha).165 E .33 -.165(ve r)-.22 H
(educed the rate).165 E
(of error losses to under 1%, making their impact mar)100.8 397 Q
(ginal [CR04].)-.198 E/F1 11/Times-Bold@0 SF(15.2.)72 423 Q/F2 13
/Times-Bold@0 SF -.13(Pa)5.5 G(ck).13 E(et Reordering)-.13 E F0 2.75
(De\214nition: Due)100.8 449 R(to the connectionless nature of IP)2.75 E
2.75(,p)-1.221 G(ack)-2.75 E(ets can arri)-.11 E .33 -.165(ve o)-.275 H
(ut of order at their).165 E 2.75(destination. P)100.8 462 R(ack)-.165 E
(et reordering e)-.11 E -.165(ve)-.275 G(nts can occur at v).165 E
(arying times and to v)-.275 E(arying de)-.275 E(grees.)-.165 E -.165
(Fo)100.8 475 S 2.75(re).165 G
(xample, a particular channel may reorder one out of ten pack)-2.915 E
(ets and the reordered)-.11 E(pack)100.8 488 Q(et arri)-.11 E -.165(ve)
-.275 G 2.75(st).165 G(hree pack)-2.75 E(ets out of order)-.11 E(.)-.605
E(Determining F)100.8 514 Q 2.75(actors: F)-.165 F
(or the most part, pack)-.165 E
(et reordering on wireless links rarely occurs.)-.11 E(Ho)100.8 527 Q
(we)-.275 E -.165(ve)-.275 G .88 -.44(r, p).165 H(ack).44 E
(et re-ordering can occur due to link layer error reco)-.11 E -.165(ve)
-.165 G(ry).165 E 5.5(.E)-.715 G(xtensi)-5.5 E .33 -.165(ve p)-.275 H
(ack).165 E(et)-.11 E(reordering has been sho)100.8 540 Q
(wn to occur with particular handof)-.275 E 2.75(fm)-.275 G
(echanisms, and is de\214nitely)-2.75 E
(detrimental to transport performance [GF04].)100.8 553 Q(Ef)100.8 579 Q
(fects on congestion control metrics:)-.275 E -.44(Wi)5.5 G(th TCP).44 E
2.75(,p)-1.221 G(ack)-2.75 E(et reordering can cause the recei)-.11 E
-.165(ve)-.275 G(r).165 E(to w)100.8 592 Q(ait for the arri)-.11 E -.275
(va)-.275 G 2.75(lo).275 G 2.75(fp)-2.75 G(ack)-2.75 E
(ets that are out of order)-.11 E 2.75(,s)-.44 G(ince the recei)-2.75 E
-.165(ve)-.275 G 2.75(rm).165 G(ust reassemble)-2.75 E(the pack)100.8
605 Q
(ets in the correct order before passing them up to the application.)
-.11 E -.44(Wi)5.5 G(th TCP and).44 E(other transport protocols, pack)
100.8 618 Q(et reordering can also result in the sender incorrectly)-.11
E(inferring pack)100.8 631 Q(et loss, triggering pack)-.11 E
(et retransmissions and congestion control responses.)-.11 E
(Measurements: Measurements by Zhou and Mie)100.8 657 Q(ghem sho)-.165 E
2.75(wt)-.275 G(hat reordering happens quite)-2.75 E
(often in the Internet, b)100.8 670 Q(ut fe)-.22 E 2.75(ws)-.275 G
(treams ha)-2.75 E .33 -.165(ve m)-.22 H(ore than tw).165 E 2.75(or)-.11
G(eordered pack)-2.75 E(ets [ZM04].)-.11 E -.165(Fo)5.5 G(r).165 E
(further measurements, see [ANP06][BPS99][LC05].)100.8 683 Q(Flo)72 769
Q(yd, K)-.11 E 262.479(ohler Section)-.385 F 2.75(15.2. [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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E/F1 11/Times-Bold@0 SF(15.3.)72
85 Q/F2 13/Times-Bold@0 SF(Delay V)5.5 E(ariation)-1.196 E F0 2.75
(De\214nition: Delay)100.8 111 R -1.221(Va)2.75 G
(riation occurs when selected pack)1.221 E(ets of a gi)-.11 E -.165(ve)
-.275 G 2.75<6e8d>.165 G .55 -.275(ow e)-2.75 H(xperience a).11 E(dif)
100.8 124 Q(ference in the One-W)-.275 E(ay-Delay across a netw)-.88 E
2.75(ork. Delay)-.11 F -.275(va)2.75 G(riation can be caused by a).275 E
-.275(va)100.8 137 S(riation in propag).275 E
(ation, transmission and queueing delay that can occur across links or)
-.055 E(netw)100.8 150 Q(ork nodes.)-.11 E(Determining F)100.8 176 Q
2.75(actors: Delay)-.165 F(and delay v)2.75 E
(ariation is introduced due to v)-.275 E(arious features of)-.275 E
(wireless links [IMLGK03].)100.8 189 Q(The delay e)5.5 E
(xperience by subsequent pack)-.165 E(ets of a gi)-.11 E -.165(ve)-.275
G 2.75<6e8d>.165 G -.275(ow)-2.75 G(can change due to On-Demand Resourc\
e Allocation, which allocates a wireless channel to)100.8 202 Q 2.75(au)
100.8 215 S(ser based on current bandwidth a)-2.75 E -.275(va)-.22 G
(ilability).275 E 5.5(.I)-.715 G 2.75(na)-5.5 G
(ddition, FEC and ARQ, which are)-2.75 E(commonly used to combat error \
loss on wireless links, can introduce delay into the)100.8 228 Q
(channel, depending on the de)100.8 241 Q
(gree of error loss that occurs.)-.165 E(These mechanisms either)5.5 E
(resend pack)100.8 254 Q(ets that ha)-.11 E .33 -.165(ve b)-.22 H
(een corrupted or attempt to reco).165 E -.165(ve)-.165 G 2.75(rt).165 G
(he actual corrupted pack)-2.75 E(et,)-.11 E
(which both add delay to the channel.)100.8 267 Q(Ef)100.8 293 Q
(fect on congestion control metrics:)-.275 E 2.75(As)5.5 G(pik)-2.75 E
2.75(ei)-.11 G 2.75(nd)-2.75 G(elay can ha)-2.75 E .33 -.165(ve a n)-.22
H -2.475 -.165(eg a).165 H(ti).165 E .33 -.165(ve i)-.275 H(mpact on)
.165 E(transport protocols [AAR03][A)100.8 306 Q(GR00][CR04]. T)-.44 E
(ransport protocols use timers for loss)-.385 E(reco)100.8 319 Q -.165
(ve)-.165 G
(ry and for congestion control, which are set according to the R).165 E
(TT)-.66 E 5.5(.D)-.814 G(elay spik)-5.5 E(es can)-.11 E(trigger spurio\
us timeouts that cause unnecessary retransmissions and incorrect conges\
tion)100.8 332 Q(control responses.)100.8 345 Q(if these delay spik)5.5
E(es continue, the)-.11 E 2.75(yc)-.165 G
(an in\215ate the retransmission)-2.75 E(timeout, increasing the w)100.8
358 Q(ait before a dropped pack)-.11 E(et is reco)-.11 E -.165(ve)-.165
G 2.75(red. Delay-based).165 F(congestion)2.75 E
(control mechanisms \(e.g. TCP V)100.8 371 Q -2.475 -.165(eg a)-1.221 H
(s, TCP W).165 E(estw)-.88 E
(ood, etc.\) use end-to-end delay to control)-.11 E
(the sending rate of the sender)100.8 384 Q 5.5(.D)-.605 G
(elay-based congestion control mechanisms use delay to)-5.5 E
(indicate when there is congestion in the netw)100.8 397 Q 2.75
(ork. When)-.11 F(delay v)2.75 E(ariation occurs for reasons)-.275 E
(other than queueing delay)100.8 410 Q 2.75(,d)-.715 G
(elay based congestion control mechanisms can reduce the)-2.75 E
(sending rate unnecessarily)100.8 423 Q 5.5(.R)-.715 G
(ate-based protocols can perform poorly as the)-5.5 E 2.75(yd)-.165 G
2.75(on)-2.75 G(ot adjust)-2.75 E
(the sending rate after a change in the R)100.8 436 Q(TT)-.66 E 2.75(,p)
-.814 G(ossibly creating unnecessary congestion)-2.75 E([GF04].)100.8
449 Q 2.75(Measurements: Cellular)100.8 475 R
(links, particularly GPRS and CDMA2000, can ha)2.75 E .33 -.165(ve o)
-.22 H(ne-w).165 E(ay)-.11 E(latencies v)100.8 488 Q
(arying from 100 to 500 ms [IMLGK03].)-.275 E(The length of a delay v)
5.5 E(ariation e)-.275 E -.165(ve)-.275 G(nt).165 E(can v)100.8 501 Q
(ary from three to \214fteen seconds and the frequenc)-.275 E 2.75(ya)
-.165 G 2.75(tw)-2.75 G(hich delay v)-2.75 E(ariation e)-.275 E -.165
(ve)-.275 G(nts).165 E(occur can be an)100.8 514 Q
(ywhere from 40 to 400 seconds.)-.165 E
(GEO satellite links tend not see much)5.5 E -.275(va)100.8 527 S
(riation in delay).275 E 2.75(,w)-.715 G
(hile LEO satellite links can see signi\214cant v)-2.75 E
(ariability in delay due to)-.275 E
(the constant motion of satellites and multiple hops.)100.8 540 Q
(The delay v)5.5 E(ariation of LEO satellite)-.275 E
(links can be from 40 to 400ms [GK98].)100.8 553 Q F1(15.4.)72 579 Q F2
(Band)5.5 E(width V)-.195 E(ariation)-1.196 E F0 2.75(De\214nition: The)
100.8 605 R(bandwidth of a wireless channel can v)2.75 E(ary o)-.275 E
-.165(ve)-.165 G 2.75(rt).165 G(ime during a single)-2.75 E
(session, as wireless netw)100.8 618 Q(orks can change the a)-.11 E
-.275(va)-.22 G(ilable bandwidth allotted to a user).275 E(.)-.605 E
(Therefore, a user may ha)100.8 631 Q .33 -.165(ve a l)-.22 H -.275(ow)
.165 G(-bandwidth channel for part of their session and a high-).275 E
(bandwidth channel the remainder of the session.)100.8 644 Q
(The bandwidth of the channel can v)5.5 E(ary)-.275 E
(abruptly or gradually)100.8 657 Q 2.75(,a)-.715 G 2.75(tv)-2.75 G
(arious interv)-3.025 E(als, and these v)-.275 E
(ariations can occur at dif)-.275 E(ferent times.)-.275 E(On-demand Res\
ource Allocation is one of the mechanisms used to dynamically allocate)
100.8 670 Q(resources to users according to system load and traf)100.8
683 Q(\214c priority)-.275 E 5.5(.F)-.715 G(or instance, in GPRS a)
-5.665 E(radio channel is allocated when data arri)100.8 696 Q -.165(ve)
-.275 G 2.75(st).165 G -2.31 -.275(ow a)-2.75 H(rd the user).275 E 2.75
(,a)-.44 G(nd released when the queue)-2.75 E(size f)100.8 709 Q
(alls belo)-.11 E 2.75(wac)-.275 G(ertain threshold [GP)-2.75 E
(AR02][W01].)-1.012 E(Flo)72 769 Q(yd, K)-.11 E 262.479(ohler Section)
-.385 F 2.75(15.4. [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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E(Determining F)100.8 85 Q 2.75
(actors: The)-.165 F(amount of bandwidth of a gi)2.75 E -.165(ve)-.275 G
2.75(nc).165 G(hannel that is allocated by the)-2.75 E(wireless netw)
100.8 98 Q(ork to a user can v)-.11 E(ary based on a number of f)-.275 E
2.75(actors. F)-.11 F(actors such as wireless)-.165 E(conditions or the\
 amount of users connected to the base station both af)100.8 111 Q
(fect the a)-.275 E -.275(va)-.22 G(ilable).275 E(capacity [IMLGK03].)
100.8 124 Q
(In certain satellite systems, technologies such as On-Demand)5.5 E
(Resource Allocation constantly adjust the a)100.8 137 Q -.275(va)-.22 G
(ilable bandwidth for a gi).275 E -.165(ve)-.275 G 2.75(nu).165 G(ser e)
-2.75 E -.165(ve)-.275 G(ry).165 E
(second, which can cause signi\214cant bandwidth v)100.8 150 Q
(ariation during a session.)-.275 E(On-Demand)5.5 E
(Resource Allocation is designed into most 2.5 and 3G wireless netw)
100.8 163 Q(orks, b)-.11 E(ut it can be)-.22 E(implemented dif)100.8 176
Q(ferently from netw)-.275 E(ork to netw)-.11 E(ork, resulting in dif)
-.11 E(ferent impacts on the)-.275 E(link.)100.8 189 Q(Ef)100.8 215 Q
(fect on congestion control metrics:)-.275 E
(In the absence of congestion, congestion control)5.5 E
(mechanisms increase the sending rate gradually o)100.8 228 Q -.165(ve)
-.165 G 2.75(rm).165 G(ultiple round-trip times.)-2.75 E(If the)5.5 E(b\
andwidth of a wireless channel suddenly increases and this increases th\
e bandwidth)100.8 241 Q -.22(av)100.8 254 S(ailable on the end-to-end p\
ath, the transport protocol might not be able to increase its)-.055 E
(sending rate quickly enough to use the ne)100.8 267 Q(wly-a)-.275 E
-.275(va)-.22 G(ilable bandwidth [24].).275 E(If the bandwidth)5.5 E(of\
 a wireless channel suddenly decreases and this decreases the bandwidth\
 a)100.8 280 Q -.275(va)-.22 G(ilable on the).275 E(end-to-end path, th\
e sender might not decrease its sending rate quickly enough, resulting \
in)100.8 293 Q(transient congestion.)100.8 306 Q
(Frequent changes of bandwidth on a wireless channel can result in)5.5 E
(the a)100.8 319 Q -.165(ve)-.22 G(rage transmission rate of the channe\
l being limited by the amount of bandwidth).165 E -.22(av)100.8 332 S
(ailable during times where the channel has the lo)-.055 E
(west bandwidth.)-.275 E(Persistent delay)5.5 E -.275(va)100.8 345 S
(riation can in\215ate the retransmission timeout, increasing the w).275
E(ait before a dropped)-.11 E(pack)100.8 358 Q(et is reco)-.11 E -.165
(ve)-.165 G(red, ultimately leading to channel underutilization [GP).165
E(AR02][W01].)-1.012 E(Measurements: Further references on the measurem\
ents for the amount of bandwidth)100.8 384 Q -.275(va)100.8 397 S
(riation are needed.).275 E
(On-demand channel allocation can be modeled by introducing an)5.5 E
(additional delay when a pack)100.8 410 Q(et arri)-.11 E -.165(ve)-.275
G 2.75(st).165 G 2.75(oaq)-2.75 G
(ueue that has been empty longer than the)-2.75 E
(channel hold time \(i.e., propag)100.8 423 Q
(ation delay\). The delay v)-.055 E(alue represents the channel)-.275 E
(allocation delay)100.8 436 Q 2.75(,a)-.715 G
(nd the hold time represents the duration of channel holding after)-2.75
E(transmitting a data pack)100.8 449 Q(et [GP)-.11 E 2.75
(AR02][W01]. See)-1.012 F(also [NM01].)2.75 E/F1 11/Times-Bold@0 SF
(15.5.)72 475 Q/F2 13/Times-Bold@0 SF(Band)5.5 E
(width and Latency Asymmetry)-.195 E F0 2.75(De\214nition: The)100.8 501
R(bandwidth in the forw)2.75 E(ard direction or uplink can be dif)-.11 E
(ferent than the)-.275 E(bandwidth in the re)100.8 514 Q -.165(ve)-.275
G(rse direction or do).165 E 2.75(wnlink. Similar)-.275 F
(to bandwidth asymmetry)2.75 E 2.75(,l)-.715 G(atenc)-2.75 E(y)-.165 E
(in the forw)100.8 527 Q(ard direction or uplink can be dif)-.11 E
(ferent than latenc)-.275 E 2.75(yi)-.165 G 2.75(nt)-2.75 G(he re)-2.75
E -.165(ve)-.275 G(rse direction or).165 E(do)100.8 540 Q 2.75
(wnlink. F)-.275 F(or e)-.165 E
(xample, bandwidth asymmetry occurs in wireless netw)-.165 E
(orks where the)-.11 E(channel from the mobile to base station \(uplink\
\) has a fraction of the bandwidth of the)100.8 553 Q
(channel from the base station to the mobile channel \(do)100.8 566 Q
(wnlink\).)-.275 E(Determining F)100.8 592 Q 2.75(actors: Bandwidth)
-.165 F(and latenc)2.75 E 2.75(ya)-.165 G(symmetry can occur for a v)
-2.75 E(ariety of)-.275 E 2.75(reasons. Mobile)100.8 605 R(de)2.75 E
(vices that must transmit at lo)-.275 E(wer po)-.275 E(wer le)-.275 E
-.165(ve)-.275 G(ls to conserv).165 E 2.75(ep)-.165 G -.275(ow)-2.75 G
(er ha).275 E -.165(ve)-.22 G(lo)100.8 618 Q 2.75(wb)-.275 G
(andwidth and high latenc)-2.75 E 2.75(yt)-.165 G
(ransmission, while base stations can transmit at higher)-2.75 E(po)
100.8 631 Q(wer le)-.275 E -.165(ve)-.275 G
(ls, resulting in higher bandwidth and lo).165 E(wer latenc)-.275 E 2.75
(y[)-.165 G 2.75(IMLGK03]. In)-2.75 F(addition,)2.75 E(because applicat\
ions such as HTTP require signi\214cantly more bandwidth on the do)100.8
644 Q(wnlink)-.275 E(as opposed to the uplink, wireless netw)100.8 657 Q
(orks ha)-.11 E .33 -.165(ve b)-.22 H(een designed with asymmetry to)
.165 E(accommodate these applications.)100.8 670 Q
(Coupled with these design constraints, the en)5.5 E(vironmental)-.44 E
(conditions can add increased asymmetry [HK99].)100.8 683 Q(Ef)100.8 709
Q(fect on congestion control metrics:)-.275 E(TCP')5.5 E 2.75(sc)-.605 G
(ongestion control algorithms rely on A)-2.75 E(CK-)-.44 E
(clocking, with the reception of A)100.8 722 Q
(CKs controlling sending rates.)-.44 E(If A)5.5 E(CKs are dropped or)
-.44 E(Flo)72 769 Q(yd, K)-.11 E 262.479(ohler Section)-.385 F 2.75
(15.5. [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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E(delayed in the re)100.8 85 Q
-.165(ve)-.275 G(rse direction, then the sending rate in the forw).165 E
(ard direction can be)-.11 E 2.75(reduced. In)100.8 98 R(addition, e)
2.75 E(xcessi)-.165 E .33 -.165(ve d)-.275 H
(elay can result in a retransmit timeout and a).165 E
(corresponding reduction in the sending rate [HK99][23].)100.8 111 Q
2.75(Measurements: F)100.8 137 R(or cellular netw)-.165 E(orks the do)
-.11 E(wnlink bandwidth typically does not e)-.275 E(xceed)-.165 E
(three to six times the uplink bandwidth [IMLGK03].)100.8 150 Q(Ho)5.5 E
(we)-.275 E -.165(ve)-.275 G .88 -.44(r, d).165 H(if).44 E
(ferent cellular netw)-.275 E(orks)-.11 E
(\(e.g. IS-95, CDMA2000, etc.\) ha)100.8 163 Q .33 -.165(ve d)-.22 H(if)
.165 E(ferent ratios of bandwidth and latenc)-.275 E 2.75(ya)-.165 G
(symmetry)-2.75 E(.)-.715 E/F1 11/Times-Bold@0 SF(15.6.)72 189 Q/F2 13
/Times-Bold@0 SF(Queue Management Mechanisms)5.5 E F0(In wireless netw)
100.8 215 Q
(orks, queueing delay typically occurs at the end points of a wireless)
-.11 E(connection \(i.e. mobile de)100.8 228 Q
(vice, base station\) [GF04].)-.275 E(Measurements: F)100.8 254 Q
(or current cellular and WLAN links, the queue can plausibly be modeled)
-.165 E(with Drop-T)100.8 267 Q
(ail queueing with a con\214gurable maximum size in pack)-.88 E 2.75
(ets. The)-.11 F(use of RED)2.75 E(may be more appropriate for modeling\
 satellite or future cellular and WLAN links [GF04].)100.8 280 Q F1(16.)
72 306 Q/F3 14/Times-Bold@0 SF(Netw)5.5 E
(ork Changes Affecting Congestion)-.14 E F0(Changes in the netw)100.8
332 Q(ork can ha)-.11 E .33 -.165(ve a s)-.22 H
(igni\214cant impact on the performance of congestion).165 E
(control algorithms.)100.8 345 Q(These changes can include e)5.5 E -.165
(ve)-.275 G(nts such as the unnecessary duplication).165 E(of pack)100.8
358 Q(ets, topology changes due to node mobility)-.11 E 2.75(,a)-.715 G
(nd temporary link disconnections.)-2.75 E(These types of netw)100.8 371
Q(ork changes can be broadly cate)-.11 E
(gorized as routing changes, link)-.165 E
(disconnections and intermittent link connecti)100.8 384 Q(vity)-.275 E
2.75(,a)-.715 G(nd mobility)-2.75 E(.)-.715 E F1(16.1.)72 410 Q F2
(Routing Changes:)5.5 E(Routing Loops)6.5 E F0
(De\214nition: A routing loop is a netw)100.8 436 Q(ork e)-.11 E -.165
(ve)-.275 G(nt in which pack).165 E(ets continue to be routed in an)-.11
E(endless circle until the pack)100.8 449 Q(ets are e)-.11 E -.165(ve)
-.275 G(ntually dropped [P96].).165 E(Determining f)100.8 475 Q
(actors: Routing loops can occur when the netw)-.11 E(ork e)-.11 E
(xperiences a change in)-.165 E(connecti)100.8 488 Q
(vity which is not immediately propag)-.275 E
(ated to all of the routers [H95].)-.055 E(Netw)5.5 E(ork and)-.11 E
(node mobility are e)100.8 501 Q(xamples of netw)-.165 E(ork e)-.11 E
-.165(ve)-.275 G(nts that can cause a change in connecti).165 E(vity)
-.275 E(.)-.715 E(Ef)100.8 527 Q(fect on Congestion Control: Loops can \
rapidly lead to congestion, as a router will route)-.275 E(pack)100.8
540 Q(ets to)-.11 E -.11(wa)-.275 G(rds a destination, b).11 E
(ut the forw)-.22 E(arded pack)-.11 E
(ets end up being routed back to the)-.11 E(router)100.8 553 Q 2.75(.F)
-.605 G(urthermore, netw)-2.75 E
(ork congestion due to a routing loop with multicast pack)-.11 E
(ets will)-.11 E(be more se)100.8 566 Q -.165(ve)-.275 G
(re than with unicast pack).165 E
(ets because each router replicates multicast pack)-.11 E(ets)-.11 E
(thereby causing congestion more rapidly [P96].)100.8 579 Q(Measurement\
s: Routing dynamics that lead to temporary route loss or forw)100.8 605
Q(arding loops are)-.11 E(also called routing f)100.8 618 Q 2.75
(ailures. ICMP)-.11 F
(response messages, measured by traceroutes and pings,)2.75 E
(can be used to identify routing f)100.8 631 Q(ailures [WMWGB06].)-.11 E
F1(16.2.)72 657 Q F2(Routing Changes:)5.5 E(Fluttering)6.5 E F0(De\214n\
ition: The term \215uttering is used to describe rapidly-oscillating ro\
uting.)100.8 683 Q(Fluttering)5.5 E
(occurs when a router alternates between multiple ne)100.8 696 Q
(xt-hop routers in order to split the load)-.165 E
(among the links to those routers.)100.8 709 Q
(While \215uttering can pro)5.5 E(vide bene\214ts as a w)-.165 E
(ay to balance)-.11 E(load in a netw)100.8 722 Q
(ork, it also creates problems for TCP [P96][AP99].)-.11 E(Flo)72 769 Q
(yd, K)-.11 E 262.479(ohler Section)-.385 F 2.75(16.2. [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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E(Determining f)100.8 85 Q(actor\
s: Multi-path routing in the Internet can cause route \215uttering.)-.11
E(Route)5.5 E
(\215uttering can result in signi\214cant out-of-order pack)100.8 98 Q
(et deli)-.11 E -.165(ve)-.275 G(ry and/or frequent abrupt end-).165 E
(to-end R)100.8 111 Q(TT v)-.66 E(ariation [P97].)-.275 E(Ef)100.8 137 Q
(fect on Congestion Control Metrics: When tw)-.275 E 2.75(or)-.11 G
(outes ha)-2.75 E .33 -.165(ve d)-.22 H(if).165 E(ferent propag)-.275 E
(ation delays,)-.055 E(pack)100.8 150 Q(ets will often arri)-.11 E .33
-.165(ve a)-.275 H 2.75(tt).165 G(he destination out-of-order)-2.75 E
2.75(,d)-.44 G(epending on whether the)-2.75 E 2.75(ya)-.165 G(rri)-2.75
E -.165(ve)-.275 G(d).165 E(via the shorter route or the longer route.)
100.8 163 Q(Whene)5.5 E -.165(ve)-.275 G 2.75(raT).165 G
(CP endpoint recei)-2.75 E -.165(ve)-.275 G 2.75(sa).165 G 2.75(no)-2.75
G(ut-of-)-2.75 E(order pack)100.8 176 Q
(et, this triggers the transmission of a duplicate ackno)-.11 E
(wledgement to inform the)-.275 E(sender that the recei)100.8 189 Q
-.165(ve)-.275 G 2.75(rh).165 G
(as a hole in its sequence space. If three out-of-order pack)-2.75 E
(ets arri)-.11 E -.165(ve)-.275 G(in a ro)100.8 202 Q 1.43 -.715(w, t)
-.275 H(hen the recei).715 E -.165(ve)-.275 G 2.75(rw).165 G
(ill generate three duplicate ackno)-2.75 E(wledgements for the se)-.275
E(gment)-.165 E(that w)100.8 215 Q(as not recei)-.11 E -.165(ve)-.275 G
2.75(d. These).165 F(duplicate ackno)2.75 E(wledgements will trigger f)
-.275 E(ast retransmit by)-.11 E(the sender)100.8 228 Q 2.75(,l)-.44 G(\
eading it to reduce the sending rate and needlessly retransmit data. Th\
us, out-)-2.75 E(of-order deli)100.8 241 Q -.165(ve)-.275 G(ry can resu\
lt in unnecessary reductions in the sending rate and also in).165 E
(redundant netw)100.8 254 Q(ork traf)-.11 E(\214c, due to e)-.275 E
(xtra ackno)-.165 E(wledgements and possibly unnecessary data)-.275 E
(retransmissions [P96][AP99].)100.8 267 Q(Measurements: T)100.8 293 Q
.22 -.11(wo m)-.88 H(etrics [WMWGB06] can be used to measure the de).11
E(gree of out-of-)-.165 E(order deli)100.8 306 Q -.165(ve)-.275 G
(ry: the number of reordering pack).165 E(ets and the reordering of)-.11
E(fset. The number of)-.275 E(reordering pack)100.8 319 Q
(ets is simply the number of pack)-.11 E
(ets that are considered out of order)-.11 E 2.75(.T)-.605 G(he)-2.75 E
(reordering of)100.8 332 Q(fset for an out-of-order pack)-.275 E
(et is the dif)-.11 E(ference between the actual arri)-.275 E -.275(va)
-.275 G(l).275 E(order and the e)100.8 345 Q(xpected arri)-.165 E -.275
(va)-.275 G 2.75(lo).275 G(rder)-2.75 E 5.5(.S)-.605 G(ee also [LMJ96].)
-5.5 E/F1 11/Times-Bold@0 SF(16.3.)72 371 Q/F2 13/Times-Bold@0 SF
(Routing Changes:)5.5 E(Routing Asymmetry)6.5 E F0
(De\214nition: Routing asymmetry occurs when pack)100.8 397 Q(ets tra)
-.11 E -.165(ve)-.22 G(ling between tw).165 E 2.75(oe)-.11 G(nd-points)
-2.75 E(follo)100.8 410 Q 2.75(wd)-.275 G(if)-2.75 E
(ferent routes in the forw)-.275 E(ard and re)-.11 E -.165(ve)-.275 G
(rse directions.).165 E(The tw)5.5 E 2.75(or)-.11 G(outes could ha)-2.75
E -.165(ve)-.22 G(dif)100.8 423 Q
(ferent characteristics in terms of bandwidth, delay)-.275 E 2.75(,l)
-.715 G -2.365 -.275(ev e)-2.75 H(ls of congestion, etc. [P96].).275 E
(Determining f)100.8 449 Q
(actors: Some of the main causes for asymmetry are polic)-.11 E 2.75(yr)
-.165 G(outing, traf)-2.75 E<8c63>-.275 E(engineering, and the absence \
of a unique shortest path between a pair of hosts. While the)100.8 462 Q
(lack of a unique shortest path is one potential contrib)100.8 475 Q
(utor to asymmetric routing within)-.22 E(domains, the principal source\
 of asymmetries in backbone routers is polic)100.8 488 Q 2.75(yr)-.165 G
(outing.)-2.75 E(Another cause of routing asymmetry is adapti)100.8 501
Q .33 -.165(ve r)-.275 H(outing, in which a router shifts traf).165 E
<8c63>-.275 E(from a highly loaded link to a less loaded one, or load b\
alances across multiple paths)100.8 514 Q([P96].)100.8 527 Q(Ef)100.8
553 Q(fect on Congestion Control:)-.275 E
(When delay-based congestion control is used, asymmetry)5.5 E
(can introduce problems in estimating the one-w)100.8 566 Q(ay latenc)
-.11 E 2.75(yb)-.165 G(etween hosts.)-2.75 E 2.75(Measurements: Further)
100.8 592 R(references are needed.)2.75 E F1(16.4.)72 618 Q F2
(Link Disconnections and Intermittent Link Connecti)5.5 E(vity)-.13 E F0
(De\214nition: A link disconnection is a period when the link loses all\
 frames, until the link is)100.8 644 Q 2.75(restored. Intermittent)100.8
657 R(Link Connecti)2.75 E(vity occurs when the link is disconnected re)
-.275 E(gularly)-.165 E(and for short periods of time.)100.8 670 Q
(This is a common characteristic of wireless links,)5.5 E
(particularly those with highly mobile nodes [AP99].)100.8 683 Q
(Determining f)100.8 709 Q(actors: In a wireless en)-.11 E
(vironment, link disconnections and intermittent link)-.44 E(connecti)
100.8 722 Q(vity could occur when a mobile de)-.275 E(vice lea)-.275 E
-.165(ve)-.22 G 2.75(st).165 G(he range of a base station, which)-2.75 E
(Flo)72 769 Q(yd, K)-.11 E 262.479(ohler Section)-.385 F 2.75(16.4. [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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E(can lead to signal de)100.8 85
Q(gradation or f)-.165 E(ailure in a handof)-.11 E 2.75(f[)-.275 G
(AP99].)-2.75 E(Ef)100.8 111 Q(fect on Congestion Control Metrics: If a\
 link disconnection lasts longer than the TCP)-.275 E -1.628 -.66(RT O)
100.8 124 T 2.75(,a).66 G(nd results in a path disconnection for that p\
eriod of time, the TCP sender will)-2.75 E
(perform a retransmit timeout, resending a pack)100.8 137 Q
(et and reducing the sending rate.)-.11 E(TCP will)5.5 E(continue this \
pattern, with longer and longer retransmit timeouts, up to a retry limi\
t, until)100.8 150 Q(an ackno)100.8 163 Q(wledgement is recei)-.275 E
-.165(ve)-.275 G 2.75(d. TCP).165 F(only determines that connecti)2.75 E
(vity has been restored)-.275 E
(after a \(possibly long\) retransmit timeout follo)100.8 176 Q
(wed by the successful receipt of an A)-.275 E(CK.)-.44 E(Thus a link d\
isconnection can result in a long delay in sending accompanied by a)
100.8 189 Q(signi\214cant reduction in the sending rate [AP99].)100.8
202 Q(Measurements: End-to-end performance under realistic topology and\
 routing policies can)100.8 228 Q
(be studied; [WMWGB06] suggests controlling routing e)100.8 241 Q -.165
(ve)-.275 G(nts by injecting well-designed).165 E
(routing updates at kno)100.8 254 Q(wn times to emulate link f)-.275 E
(ailures and repairs.)-.11 E/F1 11/Times-Bold@0 SF(16.5.)72 280 Q/F2 13
/Times-Bold@0 SF(Changes in W)5.5 E(ir)-.234 E(eless Links: Mobility)
-.234 E F0(De\214nition: Netw)100.8 306 Q(ork and node mobility)-.11 E
2.75(,b)-.715 G(oth wired and wireless, allo)-2.75 E
(ws users to roam from)-.275 E(one netw)100.8 319 Q
(ork to another seamlessly without losing service.)-.11 E(Determining f)
100.8 345 Q(actors: Mobility is a k)-.11 E .33 -.165(ey a)-.11 H(ttrib)
.165 E(ute of wireless netw)-.22 E 2.75(orks. Mobility)-.11 F(can)2.75 E
(determined by the presence of intersystem hando)100.8 358 Q -.165(ve)
-.165 G(rs, an intrinsic property of most).165 E(wireless links [HS03].)
100.8 371 Q(Ef)100.8 397 Q(fect on Congestion Control Metrics: Mobility\
 presents a major challenge to transport)-.275 E
(protocols through the pack)100.8 410 Q
(et losses and delay introduced by hando)-.11 E -.165(ve)-.165 G
(rs. In addition to delay).165 E(and losses, hando)100.8 423 Q -.165(ve)
-.165 G
(rs can also cause a signi\214cant change in link bandwidth and latenc)
.165 E -.715(y.)-.165 G(Host mobility increases pack)100.8 436 Q
(et delay and delay v)-.11 E(ariation, and also de)-.275 E
(grades the throughput)-.165 E(of TCP connections in wireless en)100.8
449 Q 2.75(vironments. Also,)-.44 F(in the e)2.75 E -.165(ve)-.275 G
(nt of a handof).165 E(f, slo)-.275 E(wly-)-.275 E(responsi)100.8 462 Q
.33 -.165(ve c)-.275 H
(ongestion control can require considerable time to adapt to changes. F)
.165 E(or)-.165 E -.165(ex)100.8 475 S(ample a \215o).165 E 2.75(wu)
-.275 G(nder)-2.75 E(-utilizes a f)-.22 E(ast link after a hando)-.11 E
-.165(ve)-.165 G 2.75(rf).165 G(rom a slo)-2.75 E 2.75(wl)-.275 G
(ink [HS03].)-2.75 E 2.75(Measurements: Further)100.8 501 R
(references are needed to specify ho)2.75 E 2.75(wm)-.275 G
(obility is actually measured)-2.75 E([JEAS03].)100.8 514 Q F1(17.)72
540 Q/F3 14/Times-Bold@0 SF(Using the T)5.5 E(ools Pr)-1.288 E
(esented in this Document)-.252 E F0([T)100.8 556.6 Q 2.75(ob)-.88 G
2.75(ed)-2.75 G(one.])-2.75 E F1(18.)72 582.6 Q F3(Related W)5.5 E(ork)
-1.05 E F0([Cite "On the Ef)100.8 599.2 Q(fecti)-.275 E .33 -.165(ve E)
-.275 H -.275(va).165 G(luation of TCP" by Allman and F).275 E(alk.])
-.165 E F1(19.)72 625.2 Q F3(Conclusions)5.5 E F0([T)100.8 641.8 Q 2.75
(ob)-.88 G 2.75(ed)-2.75 G(one.])-2.75 E F1(20.)72 667.8 Q F3
(Security Considerations)5.5 E F0
(There are no security considerations in this document.)100.8 684.4 Q
(Flo)72 769 Q(yd, K)-.11 E 270.729(ohler Section)-.385 F 2.75(20. [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 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E/F1 11/Times-Bold@0 SF(21.)72 85
Q/F2 14/Times-Bold@0 SF(IAN)5.5 E 3.5(AC)-.28 G(onsiderations)-3.5 E F0
(There are no IAN)100.8 101.6 Q 2.75(Ac)-.385 G
(onsiderations in this document.)-2.75 E F1(22.)72 127.6 Q F2(Ackno)5.5
E(wledgements)-.14 E F0(Thanks to Xiaoliang \(Da)100.8 144.2 Q(vid\) W)
-.22 E(ei for feedback and contrib)-.88 E(utions to this document.)-.22
E(The)5.5 E(sections on "Challenging Lo)100.8 157.2 Q
(wer Layers" and "Netw)-.275 E(ork Changes Af)-.11 E
(fecting Congestion" are)-.275 E(contrib)100.8 170.2 Q
(utions from Jasani Rohan with Julie T)-.22 E(arr)-.88 E 2.75(,T)-.44 G
(on)-3.63 E 2.75(yD)-.165 G(esimone, Christou Christos, and)-2.75 E
-1.221(Ve)100.8 183.2 S(mulapalli Archana.)1.221 E F2(Inf)72 209.2 Q
(ormati)-.35 E .28 -.14(ve R)-.14 H(efer).14 E(ences)-.252 E F0([MA)
100.8 225.8 Q(WI] M.W)-.99 E 2.75(.G)-1.012 G(roup, Ma)-2.75 E(wi w)
-.165 E(orking group traf)-.11 E(\214c archi)-.275 E -.165(ve)-.275 G
2.75(,U).165 G(RL)-2.75 E("http://tracer)129.6 238.8 Q(.csl.son)-.605 E
-.715(y.)-.165 G(jp/ma).715 E(wi/".)-.165 E([A00] M. Allman, A W)100.8
255.4 Q(eb Serv)-.88 E(er')-.165 E 2.75(sV)-.605 G(ie)-3.41 E 2.75(wo)
-.275 G 2.75(ft)-2.75 G(he T)-2.75 E(ransport Layer)-.385 E 2.75(,C)-.44
G(omputer Communication)-2.75 E(Re)129.6 268.4 Q(vie)-.275 E 1.43 -.715
(w, 3)-.275 H(0\(5\), October 200.).715 E
([AAR03] Alhussein A. Abouzeid, Sumit Ro)100.8 285 Q 1.43 -.715(y, S)
-.11 H(tochastic Modeling of TCP in Netw).715 E(orks)-.11 E
(with Abrupt Delay V)129.6 298 Q(ariations, W)-1.221 E(ireless Netw)-.44
E(orks, V)-.11 E(.9 N.5, 2003.)-1.419 E([A)100.8 314.6 Q
(GR00] M. Allman, J. Griner)-.44 E 2.75(,a)-.44 G
(nd A. Richard. TCP Beha)-2.75 E(vior in Netw)-.22 E(orks with Dynamic)
-.11 E(Propag)129.6 327.6 Q(ation Delay)-.055 E 2.75(.G)-.715 G
(lobecom 2000, No)-2.75 E -.165(ve)-.165 G(mber 2000.).165 E
([AKM04] B. Appenzeller)100.8 344.2 Q 2.75(,I)-.44 G 2.75(.K)-2.75 G
(eslassy)-3.025 E 2.75(,a)-.715 G(nd N. McK)-2.75 E(eo)-.275 E
(wn, Sizing Router Buf)-.275 E(fers,)-.275 E(SIGCOMM 2004.)129.6 357.2 Q
([AKSJ03] J. Aikat, J. Kaur)100.8 373.8 Q 2.75(,F)-.44 G
(.D. Smith, and K. Jef)-3.63 E -.11(fa)-.275 G 1.43 -.715(y, V).11 H
(ariability in TCP Roundtrip T)-.506 E(imes,)-.385 E -.44(AC)129.6 386.8
S 2.75(MS).44 G
(IGCOMM Internet Measurement Conference, Maimi, FL, October 2003, pp.)
-2.75 E(279-284.)129.6 399.8 Q([ANP06] On Monitoring of End-to-End P)
100.8 416.4 Q(ack)-.165 E(et Reordering o)-.11 E -.165(ve)-.165 G 2.75
(rt).165 G(he Internet, B. Y)-2.75 E(e, A. P)-1.1 E(.)-1.221 E
(Jayasumana, and N. M. Piratla, 2006.)129.6 429.4 Q
([AP99] M. Allman and V)100.8 446 Q 2.75(.P)-1.419 G
(axson. On Estimating End-to-end Netw)-2.915 E(ork P)-.11 E
(ath Properties.)-.165 E(SIGCOMM, September 1999.)129.6 459 Q([BH02] F)
100.8 475.6 Q 2.75(.B)-.88 G(accelli and D. Hong, AIMD, F)-2.75 E
(airness and Fractal Scaling of TCP T)-.165 E(raf)-.385 E(\214c,)-.275 E
(Infocom 2002.)129.6 488.6 Q([B)100.8 505.2 Q(A05] E. Blanton and M. Al\
lman, "On the Impack to Bursting on TCP Performance",)-.385 E -.165(Pa)
129.6 518.2 S(ssi).165 E .33 -.165(ve a)-.275 H(nd Acti).165 E .33 -.165
(ve M)-.275 H(easurement W).165 E(orkshop, March 2005.)-.88 E
([BPS99] Bennett, J. C. R., P)100.8 534.8 Q
(artridge, C. and Shectman, N., "P)-.165 E(ack)-.165 E
(et Reordering is Not)-.11 E -.165(Pa)129.6 547.8 S(thological Netw).165
E(ork Beha)-.11 E(vior)-.22 E(," T)-.44 E(rans. on Netw)-.385 E
(orking IEEE/A)-.11 E(CM, Dec. 1999,)-.44 E(pp.789-798.)129.6 560.8 Q
([CBC95] C. Cunha, A. Besta)100.8 577.4 Q(vros, and M. Cro)-.22 E -.165
(ve)-.165 G(lla, "Characteristics of WWW Client-).165 E(based T)129.6
590.4 Q(races", B)-.385 E 2.75(UT)-.11 G(echnical Report B)-3.52 E
(UCS-95-010, 1995.)-.11 E([CR04] M. Chan and R. Ramjee, \(2004\) Impro)
100.8 607 Q(ving TCP/IP Performance o)-.165 E -.165(ve)-.165 G 2.75(rT)
.165 G(hird)-2.75 E(Generation W)129.6 620 Q(ireless Netw)-.44 E
(orks. IEEE Infocom 2004.)-.11 E([Dummynet] L. Rizzo, Dummynet, URL "ht\
tp://info.iet.unipi.it/~luigi/ip_dummynet/".)100.8 636.6 Q
([F02] C. J. Fraleigh, Pro)100.8 653.2 Q
(visioning Internet Backbone Netw)-.165 E(orks to Support Latenc)-.11 E
(y)-.165 E(Sensiti)129.6 666.2 Q .33 -.165(ve A)-.275 H 2.75
(pplications. PhD).165 F(thesis, Stanford Uni)2.75 E -.165(ve)-.275 G
(rsity).165 E 2.75(,D)-.715 G(epartment of Electrical)-2.75 E
(Engineering, June 2002.)129.6 679.2 Q([FJ92] S. Flo)100.8 695.8 Q
(yd and V)-.11 E 2.75(.J)-1.419 G(acobson, On T)-2.75 E(raf)-.385 E
(\214c Phase Ef)-.275 E(fects in P)-.275 E(ack)-.165 E(et-Switched Gate)
-.11 E -.11(wa)-.275 G(ys,).11 E(Internetw)129.6 708.8 Q
(orking: Research and Experience, V)-.11 E
(.3 N.3, September 1992, p.115-156.)-1.419 E(Flo)72 769 Q(yd, K)-.11 E
325.729(ohler [P)-.385 F(age 19])-.165 E 0 Cg EP
%%Page: 20 20
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E([FK02] S. Flo)100.8 85 Q
(yd and E. K)-.11 E(ohler)-.385 E 2.75(,I)-.44 G
(nternet Research Needs Better Models, Hotnets-I, October)-2.75 E(2002.)
129.6 98 Q([GF04] A. Gurto)100.8 114.6 Q 2.75(va)-.165 G(nd S. Flo)-2.75
E(yd. Modeling W)-.11 E(ireless Links for T)-.44 E
(ransport Protocols. A)-.385 E(CM)-.44 E
(CCR, 34\(2\):85-96, April 2004.)129.6 127.6 Q([GK98] B. Ga)100.8 144.2
Q(vish and J. Kalv)-.22 E
(enes. The Impact of Satellite Altitude on the Performance of)-.165 E
(LEOS based Communication Systems. W)129.6 157.2 Q(ireless Netw)-.44 E
(orks, 4\(2\):199--212, 1998.)-.11 E
([GM04] L. Grieco and S. Mascolo, Performance Ev)100.8 173.8 Q
(aluation and Comparison of)-.275 E -.88(We)129.6 186.8 S(stw).88 E
(ood+, Ne)-.11 E 2.75(wR)-.275 G(eno, and V)-2.75 E -2.475 -.165(eg a)
-1.221 H 2.75(sT).165 G(CP Congestion Control, CCR, April 2004.)-2.75 E
([GP)100.8 203.4 Q(AR02] A. Gurto)-1.012 E 1.43 -.715(v, M)-.165 H 2.75
(.P).715 G(assoja, O. Aalto, and M. Raitola. Multi-layer Protocol T)
-2.915 E(racing)-.385 E(in a GPRS Netw)129.6 216.4 Q
(ork. In Proc. of the IEEE V)-.11 E(ehicular T)-1.221 E
(echnology Conference)-.77 E(\(VTC02 F)129.6 229.4 Q(all\), Sept. 2002.)
-.165 E([H95] Huitema, C., \(1995\) Routing in the Internet. Prentice H\
all PTR, 1995.)100.8 246 Q([HK99] T)100.8 262.6 Q 2.75(.H)-.814 G
(enderson and R. Katz, \(1999\) T)-2.75 E
(ransport Protocols for Internet-Compatible)-.385 E(Satellite Netw)129.6
275.6 Q(orks. IEEE Journal on Selected Areas in Communications,)-.11 E
-1.419(Vo)5.5 G(l. 17,)1.419 E(No. 2, pp. 345-359, February 1999.)129.6
288.6 Q([HMTG01] C. Hollot, V)100.8 305.2 Q 2.75(.M)-1.419 G(isra, D. T)
-2.75 E -.275(ow)-.88 G(sle).275 E 1.43 -.715(y, a)-.165 H(nd W).715 E
2.75(.G)-1.012 G(ong, On Designing Impro)-2.75 E -.165(ve)-.165 G(d).165
E(Controllers for A)129.6 318.2 Q(QM Routers Supporting TCP Flo)-.605 E
(ws, IEEE Infocom, 2001.)-.275 E([HS03] R. Hsieh and A. Sene)100.8 334.8
Q(viratne. A Comparison of Mechanisms for Impro)-.275 E(ving Mobile)
-.165 E(IP Handof)129.6 347.8 Q 2.75(fL)-.275 G(atenc)-2.75 E 2.75(yf)
-.165 G(or End-to-end TCP)-2.75 E 5.5(.M)-1.221 G(OBICOM, Sept. 2003.)
-5.5 E([IMD01] G. Iannaccone, M. May)100.8 364.4 Q 2.75(,a)-.715 G
(nd C. Diot, Aggre)-2.75 E -.055(ga)-.165 G(te T).055 E(raf)-.385 E
(\214c Performance with Acti)-.275 E -.165(ve)-.275 G
(Queue Management and Drop From T)129.6 377.4 Q 2.75(ail. SIGCOMM)-.88 F
(Comput. Commun. Re)2.75 E -.715(v.)-.275 G(,).715 E
(31\(3\):4-13, 2001.)129.6 390.4 Q([IMLGK03] H. Inamura, G. Montene)
100.8 407 Q(gro, R. Ludwig, A. Gurto)-.165 E 2.75(va)-.165 G(nd F)-2.75
E 5.5(.K)-.88 G(ha\214zo)-5.5 E 1.43 -.715(v. T)-.165 H(CP).715 E -.165
(ove)129.6 420 S 2.75(rS).165 G
(econd \(2.5G\) and Third \(3G\) Generation W)-2.75 E(ireless Netw)-.44
E(orks. RFC 3484,)-.11 E(IETF)129.6 433 Q 2.75(,F)-.88 G(ebruary 2003.)
-2.75 E([JD02] H. Jiang and C. Do)100.8 449.6 Q(vrolis, P)-.165 E(assi)
-.165 E .33 -.165(ve E)-.275 H(stimation of TCP Round-trip T).165 E
(imes, Computer)-.385 E(Communication Re)129.6 462.6 Q(vie)-.275 E 1.43
-.715(w, 3)-.275 H(2\(3\), July 2002.).715 E
([JEAS03] A. Jardosh, E. Belding-Ro)100.8 479.2 Q(yer)-.11 E 2.75(,K)
-.44 G 2.75(.A)-2.75 G(lmeroth, and S. Suri. T)-2.75 E -2.31 -.275(ow a)
-.88 H(rds Realistic).275 E(Mobility Models for Mobile Ad Hoc Netw)129.6
492.2 Q 2.75(orks. MOBICOM,)-.11 F(Sept. 2003.)2.75 E
([LC01] J. Liu and Mark Cro)100.8 508.8 Q -.165(ve)-.165 G
(lla, Using Loss P).165 E(airs to Disco)-.165 E -.165(ve)-.165 G 2.75
(rN).165 G(etw)-2.75 E(ork Properties, A)-.11 E(CM)-.44 E
(SIGCOMM Internet Measurement W)129.6 521.8 Q(orkshop, 2001.)-.88 E
([LC05] X. Luo and R. K. C. Chang, No)100.8 538.4 Q -.165(ve)-.165 G
2.75(lA).165 G(pproaches to End-to-end P)-2.75 E(ack)-.165 E
(et Reordering)-.11 E(Measurement, 2005.)129.6 551.4 Q([LMJ96] Labo)
100.8 568 Q(vitz, C., Malan, G.R., and Jahanian, F)-.165 E
(., \(1996\) Internet Routing Instability)-.88 E(.)-.715 E
(Proceedings of SIGCOMM 96.)129.6 581 Q
([MAF05] A. Medina, M. Allman, and A. Flo)100.8 597.6 Q 2.75
(yd. Measuring)-.11 F(the Ev)2.75 E(olution of T)-.22 E(ransport)-.385 E
(Protocols in the Internet.)129.6 610.6 Q(Computer Communication Re)5.5
E(vie)-.275 E 1.43 -.715(w, A)-.275 H(pril 2005.).715 E
([NISTNet] NIST Net, URL "http://snad.ncsl.nist.go)100.8 627.2 Q
(v/itg/nistnet/".)-.165 E([NM01] J. Neale and A. Mohsen. Impact of CF-D)
100.8 643.8 Q(AMA on TCP via Satellite Performance.)-.44 E(Globecom, No)
129.6 656.8 Q 1.43 -.715(v. 2)-.165 H(001.).715 E([P96] P)100.8 673.4 Q
(axson, V)-.165 E(., \(1996\) End-to-end Routing Beha)-1.419 E
(vior in the Internet.)-.22 E(Proceedings of)5.5 E
(SIGCOMM 96, pp. 25-38, August 1992.)129.6 686.4 Q([P97] V)100.8 703 Q
2.75(.P)-1.419 G(axson. End-to-end Routing Beha)-2.915 E
(vior in the Internet. IEEE/A)-.22 E(CM T)-.44 E(ransactions on)-.385 E
(Netw)129.6 716 Q(orking, 5\(5\):60115, October 1997.)-.11 E(Flo)72 769
Q(yd, K)-.11 E 325.729(ohler [P)-.385 F(age 20])-.165 E 0 Cg EP
%%Page: 21 21
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E
([PJD04] R. Prasad, M. Jain, and C. Do)100.8 85 Q(vrolis, On the Ef)
-.165 E(fecti)-.275 E -.165(ve)-.275 G(ness of Delay-Based).165 E
(Congestion A)129.6 98 Q -.22(vo)-.814 G
(idance, PFLDnet 2004, February 2004.).22 E([QZK01] L. Qiu, Y)100.8
114.6 Q 2.75(.Z)-1.419 G(hang, and S. K)-2.75 E(esha)-.275 E 1.43 -.715
(v, U)-.22 H(nderstanding the Performance of Man).715 E 2.75(yT)-.165 G
(CP)-2.75 E(Flo)129.6 127.6 Q(ws, Comput. Netw)-.275 E
(orks, 37\(3-4\):277-306, 2001.)-.11 E([RSB01] R. Riedi, S. Sarv)100.8
144.2 Q(otham, and R. V)-.22 E(araniuk, Connection-le)-1.221 E -.165(ve)
-.275 G 2.75(lA).165 G(nalysis and)-2.75 E(Modeling of Netw)129.6 157.2
Q(ork T)-.11 E(raf)-.385 E(\214c, SIGCOMM Internet Measurement W)-.275 E
(orkshop, 2001.)-.88 E([R)100.8 173.8 Q(TW05] G. Raina, D. T)-.66 E
-.275(ow)-.88 G(sle).275 E 1.43 -.715(y, a)-.165 H(nd D. W).715 E
(ischik, Control Theory for Buf)-.44 E(fer Sizing, CCR,)-.275 E
(July 2005.)129.6 186.8 Q([LS06] D. Leith and R. Shorten, Impact of Dro\
p Synchronisation on TCP F)100.8 203.4 Q(airness in High)-.165 E
(Bandwidth-Delay Product Netw)129.6 216.4 Q(orks, Proc. Protocols for F)
-.11 E(ast Long Distance)-.165 E(Netw)129.6 229.4 Q(orks, 2006.)-.11 E
([TG] T)100.8 246 Q(raf)-.385 E(\214c Generators for Internet T)-.275 E
(raf)-.385 E(\214c W)-.275 E(eb P)-.88 E(age, URL)-.165 E("http://www)
129.6 259 Q(.icir)-.715 E(.or)-.605 E(g/models/traf)-.198 E
(\214cgenerators.html".)-.275 E([U)100.8 275.6 Q
(A01] U. of Auckland, Auckland-vi trace data, June 2001.)-.44 E(URL)5.5
E("http://w)129.6 288.6 Q(ans.cs.w)-.11 E(aikato.ac.nz/w)-.11 E
(and/wits/auck/6/".)-.11 E([UW02] UW)100.8 305.2 Q(-Madison, Netw)-.715
E(ork Performance Statistics, October 2002.)-.11 E(URL)5.5 E
("http://wwwstats.net.wisc.edu/".)129.6 318.2 Q([W01] B. W)100.8 334.8 Q
(alk)-.88 E(e. Mobile Radio Netw)-.11 E(orks, Netw)-.11 E
(orking and Protocols \(2.)-.11 E(Ed.\). W)5.5 E(ile)-.44 E 2.75(y&)
-.165 G(Sons, 2001.)129.6 347.8 Q([WM05] D. W)100.8 364.4 Q
(ischik and N. McK)-.44 E(eo)-.275 E(wn, Buf)-.275 E
(fer sizes for Core Routers, CCR, July 2005.)-.275 E
(URL "http://yuba.stanford.edu/~nickm/papers/Buf)129.6 377.4 Q
(ferSizing.pdf".)-.275 E([WMWGB06] F)100.8 394 Q 2.75(.W)-.88 G
(ang, Z. M. Mao, J. W)-3.63 E
(ang, L. Gao and R. Bush. A Measurement Study)-.88 E
(on the Impact of Routing Ev)129.6 407 Q(ents on End-to-End Internet P)
-.165 E(ath Performance,)-.165 E(SIGCOMM, 2006.)129.6 420 Q
([ZM04] X. Zhou and P)100.8 436.6 Q 2.75(.V)-1.221 G(an Mie)-3.971 E
(ghem, Reordering of IP P)-.165 E(ack)-.165 E(ets in Internet, P)-.11 E
(AM 2004.)-1.012 E([ZSC91] L. Zhang, S. Shenk)100.8 453.2 Q(er)-.11 E
2.75(,a)-.44 G(nd D.D. Clark, Observ)-2.75 E(ations and Dynamics of a)
-.275 E(Congestion Control Algorithm: the Ef)129.6 466.2 Q(fects of T)
-.275 E -.11(wo)-.88 G(-w).11 E(ay T)-.11 E(raf)-.385 E
(\214c, SIGCOMM 1991.)-.275 E([22])100.8 482.8 Q([23])100.8 499.4 Q
([24])100.8 516 Q/F1 14/Times-Bold@0 SF(Editors' Addr)72 542 Q(esses)
-.252 E F0(Sally Flo)100.8 558.6 Q(yd <\215o)-.11 E(yd@icir)-.11 E(.or)
-.605 E(g>)-.198 E(ICSI Center for Internet Research)100.8 571.6 Q
(1947 Center Street, Suite 600)100.8 584.6 Q(Berk)100.8 597.6 Q(ele)-.11
E 1.43 -.715(y, C)-.165 H 2.75(A9).715 G(4704)-2.75 E(USA)100.8 610.6 Q
(Eddie K)100.8 627.2 Q(ohler <k)-.385 E(ohler@cs.ucla.edu>)-.11 E
(4531C Boelter Hall)100.8 640.2 Q(UCLA Computer Science Department)100.8
653.2 Q(Los Angeles, CA 90095)100.8 666.2 Q(USA)100.8 679.2 Q(Flo)72 769
Q(yd, K)-.11 E 325.729(ohler [P)-.385 F(age 21])-.165 E 0 Cg EP
%%Page: 22 22
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 71.587(-DRAFT Expires:)-1.012 F
(January 2008)2.75 E(July 2007)123.726 E/F1 14/Times-Bold@0 SF
(Full Copyright Statement)72 85 Q F0(Cop)100.8 101.6 Q
(yright \(C\) The IETF T)-.11 E(rust \(2007\).)-.385 E
(This document is subject to the rights, licenses and)5.5 E
(restrictions contained in BCP 78, and e)100.8 114.6 Q
(xcept as set forth therein, the authors retain all their)-.165 E
(rights.)100.8 127.6 Q
(This document and the information contained herein are pro)100.8 144.2
Q(vided on an "AS IS" basis and)-.165 E(THE CONTRIB)100.8 157.2 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)100.8 170.2 Q 2.75(,T)
-1.419 G(HE IETF TR)-2.75 E(UST AND THE)-.44 E(INTERNET ENGINEERING T)
100.8 183.2 Q(ASK FORCE DISCLAIM ALL W)-1.023 E(ARRANTIES,)-1.32 E
(EXPRESS OR IMPLIED, INCLUDING B)100.8 196.2 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)-1.32 E(THA)
100.8 209.2 Q 2.75(TT)-1.221 G(HE USE OF THE INFORMA)-2.75 E
(TION HEREIN WILL NO)-1.221 E 2.75(TI)-.44 G(NFRINGE ANY)-2.75 E
(RIGHTS OR ANY IMPLIED W)100.8 222.2 Q(ARRANTIES OF MERCHANT)-1.32 E
(ABILITY OR FITNESS)-1.023 E(FOR A P)100.8 235.2 Q(AR)-1.012 E
(TICULAR PURPOSE.)-.66 E F1(Intellectual Pr)72 261.2 Q(operty)-.252 E F0
(The IETF tak)100.8 277.8 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 clai\
med to pertain to the implementation or use of the)100.8 290.8 Q
(technology described in this document or the e)100.8 303.8 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)100.8 316.8 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)100.8 329.8 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.)100.8 342.8 Q
(Copies of IPR disclosures made to the IETF Secretariat and an)100.8
359.4 Q 2.75(ya)-.165 G(ssurances of licenses to)-2.75 E(be made a)100.8
372.4 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)100.8 385.4 Q(speci\214cati\
on can be obtained from the IETF on-line IPR repository at)100.8 398.4 Q
(http://www)100.8 411.4 Q(.ietf.or)-.715 E(g/ipr)-.198 E(.)-.605 E
(The IETF in)100.8 428 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)100.8 441
Q -.165(ve)-.165 G 2.75(rt).165 G(echnology that may be)-2.75 E
(required to implement this standard.)100.8 454 Q
(Please address the information to the IETF at ietf-)5.5 E(ipr@ietf.or)
100.8 467 Q(g.)-.198 E(Flo)72 769 Q(yd, K)-.11 E 325.729(ohler [P)-.385
F(age 22])-.165 E 0 Cg EP
%%Trailer
end
%%EOF