One document matched: draft-ietf-rmt-bb-pgmcc-01.ps

%!PS-Adobe-3.0
%%Creator: groff version 1.17.2
%%CreationDate: Thu Jun 27 19:03:01 2002
%%DocumentNeededResources: font Courier-Bold
%%+ font Times-Bold
%%+ font Times-Roman
%%+ font Courier
%%DocumentSuppliedResources: procset grops 1.17 2
%%Pages: 17
%%PageOrder: Ascend
%%Orientation: Portrait
%%EndComments
%%BeginProlog
%%BeginResource: procset grops 1.17 2
/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
/FL{
currentgray exch setgray fill setgray
}bind def
/BL/fill load def
/LW/setlinewidth load def
/RE{
findfont
dup maxlength 1 index/FontName known not{1 add}if dict begin
{
1 index/FID ne{def}{pop pop}ifelse
}forall
/Encoding exch def
dup/FontName exch def
currentdict end definefont pop
}bind def
/DEFS 0 def
/EBEGIN{
moveto
DEFS begin
}bind def
/EEND/end load def
/CNT 0 def
/level1 0 def
/PBEGIN{
/level1 save def
translate
div 3 1 roll div exch scale
neg exch neg exch translate
0 setgray
0 setlinecap
1 setlinewidth
0 setlinejoin
10 setmiterlimit
[]0 setdash
/setstrokeadjust where{
pop
false setstrokeadjust
}if
/setoverprint where{
pop
false setoverprint
}if
newpath
/CNT countdictstack def
userdict begin
/showpage{}def
}bind def
/PEND{
clear
countdictstack CNT sub{end}repeat
level1 restore
}bind def
end def
/setpacking where{
pop
setpacking
}if
%%EndResource
%%IncludeResource: font Courier-Bold
%%IncludeResource: font Times-Bold
%%IncludeResource: font Times-Roman
%%IncludeResource: font Courier
grops begin/DEFS 1 dict def DEFS begin/u{.001 mul}bind def end/RES 72
def/PL 792 def/LS false def/ENC0[/asciicircum/asciitilde/Scaron/Zcaron
/scaron/zcaron/Ydieresis/trademark/quotesingle/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef/.notdef
/.notdef/.notdef/space/exclam/quotedbl/numbersign/dollar/percent
/ampersand/quoteright/parenleft/parenright/asterisk/plus/comma/hyphen
/period/slash/zero/one/two/three/four/five/six/seven/eight/nine/colon
/semicolon/less/equal/greater/question/at/A/B/C/D/E/F/G/H/I/J/K/L/M/N/O
/P/Q/R/S/T/U/V/W/X/Y/Z/bracketleft/backslash/bracketright/circumflex
/underscore/quoteleft/a/b/c/d/e/f/g/h/i/j/k/l/m/n/o/p/q/r/s/t/u/v/w/x/y
/z/braceleft/bar/braceright/tilde/.notdef/quotesinglbase/guillemotleft
/guillemotright/bullet/florin/fraction/perthousand/dagger/daggerdbl
/endash/emdash/ff/fi/fl/ffi/ffl/dotlessi/dotlessj/grave/hungarumlaut
/dotaccent/breve/caron/ring/ogonek/quotedblleft/quotedblright/oe/lslash
/quotedblbase/OE/Lslash/.notdef/exclamdown/cent/sterling/currency/yen
/brokenbar/section/dieresis/copyright/ordfeminine/guilsinglleft
/logicalnot/minus/registered/macron/degree/plusminus/twosuperior
/threesuperior/acute/mu/paragraph/periodcentered/cedilla/onesuperior
/ordmasculine/guilsinglright/onequarter/onehalf/threequarters
/questiondown/Agrave/Aacute/Acircumflex/Atilde/Adieresis/Aring/AE
/Ccedilla/Egrave/Eacute/Ecircumflex/Edieresis/Igrave/Iacute/Icircumflex
/Idieresis/Eth/Ntilde/Ograve/Oacute/Ocircumflex/Otilde/Odieresis
/multiply/Oslash/Ugrave/Uacute/Ucircumflex/Udieresis/Yacute/Thorn
/germandbls/agrave/aacute/acircumflex/atilde/adieresis/aring/ae/ccedilla
/egrave/eacute/ecircumflex/edieresis/igrave/iacute/icircumflex/idieresis
/eth/ntilde/ograve/oacute/ocircumflex/otilde/odieresis/divide/oslash
/ugrave/uacute/ucircumflex/udieresis/yacute/thorn/ydieresis]def
/Courier@0 ENC0/Courier RE/Times-Roman@0 ENC0/Times-Roman RE
/Times-Bold@0 ENC0/Times-Bold RE/Courier-Bold@0 ENC0/Courier-Bold RE
%%EndProlog
%%Page: 1 1
%%BeginPageSetup
BP
%%EndPageSetup
/F0 10/Courier-Bold@0 SF(Internet Engineering Task Force)72 85 Q(RMT WG)
210 E 228(INTERNET-DRAFT Luigi)72 98 R(Rizzo/U. Pisa)6 E 72
(draft-ietf-rmt-bb-pgmcc-01.ps Gianluca)72 111 R(Iannaccone/SprintLabs)6
E(Lorenzo Vicisano/Cisco)372 124 Q(Mark Handley/ICSI)402 137 Q
(27 June 2002)432 150 Q(Expires: December 2002)372 163 Q/F1 14
/Times-Bold@0 SF(PGMCC single rate multicast congestion contr)140.951
188 Q(ol:)-.252 E(Pr)222.991 201 Q(otocol Speci\214cation)-.252 E/F2 11
/Times-Bold@0 SF(Status of this Document)72 246 Q/F3 11/Times-Roman@0 SF
(This document is an Internet-Draft and is in full conformance with all\
 pro)72 262.6 Q(visions of Section 10 of)-.165 E(RFC2026.)72 275.6 Q
(Internet-Drafts are w)72 301.6 Q
(orking documents of the Internet Engineering T)-.11 E(ask F)-.88 E
(orce \(IETF\), its areas,)-.165 E(and its w)72 314.6 Q(orking groups.)
-.11 E(Note that other groups may also distrib)5.5 E(ute w)-.22 E
(orking documents as)-.11 E(Internet-Drafts.)72 327.6 Q
(Internet-Drafts are v)72 353.6 Q
(alid for a maximum of six months and may be updated, replaced, or)-.275
E(obsoleted by other documents at an)72 366.6 Q 2.75(yt)-.165 G 2.75
(ime. It)-2.75 F(is inappropriate to use Internet-Drafts as reference)
2.75 E(material or to cite them other than as a "w)72 379.6 Q
(ork in progress".)-.11 E
(The list of current Internet-Drafts can be accessed at http://www)72
405.6 Q(.ietf.or)-.715 E(g/ietf/1id-abstracts.txt)-.198 E 1.76 -.88
(To v)72 431.6 T(ie).88 E 2.75(wt)-.275 G(he list Internet-Draft Shado)
-2.75 E 2.75(wD)-.275 G(irectories, see http://www)-2.75 E(.ietf.or)
-.715 E(g/shado)-.198 E -.715(w.)-.275 G(html.).715 E
(This document is a product of the IETF RMT WG.)72 457.6 Q
(Comments should be addressed to the)5.5 E(authors, or the WG')72 470.6
Q 2.75(sm)-.605 G(ailing list at rmt@lbl.go)-2.75 E -.715(v.)-.165 G F2
(Abstract)267.534 502.6 Q F3(This document describes PGMCC, a single ra\
te multicast congestion control scheme)97 525.2 Q
(which is TCP-friendly and achie)97 538.2 Q -.165(ve)-.275 G 2.75(ss)
.165 G(calability)-2.75 E 2.75(,s)-.715 G(tability and f)-2.75 E
(ast response to v)-.11 E(ariations)-.275 E(in netw)97 551.2 Q
(ork conditions.)-.11 E
(PGMCC is suitable for both non-reliable and reliable data)5.5 E 2.75
(transfers. It)97 564.2 R(is mainly designed for N)2.75 E
(AK-based multicast protocols, and uses a)-.385 E(windo)97 577.2 Q
(w-based, TCP-lik)-.275 E 2.75(ec)-.11 G(ontrol loop using positi)-2.75
E .33 -.165(ve A)-.275 H(CKs between one representati)-.275 E -.165(ve)
-.275 G(of the recei)97 590.2 Q -.165(ve)-.275 G 2.75(rg).165 G
(roup \(the A)-2.75 E(CKER\) and the sender)-.44 E 5.5(.T)-.605 G(he A)
-5.5 E(CKER is selected)-.44 E(dynamically and may change o)97 603.2 Q
-.165(ve)-.165 G 2.75(rt).165 G(ime.)-2.75 E(PGMCC is made of tw)97
629.2 Q 2.75(oc)-.11 G(omponents: a windo)-2.75 E
(w-based control loop, which closely)-.275 E(mimics TCP beha)97 642.2 Q
(viour)-.22 E 2.75(,a)-.44 G(nd a f)-2.75 E(ast and lo)-.11 E(w-o)-.275
E -.165(ve)-.165 G(rhead procedure to select \(and track).165 E
(changes of\) the A)97 655.2 Q 2.75(CKER. The)-.44 F(scheme is rob)2.75
E(ust to measurement errors, and supports)-.22 E -.11(fa)97 668.2 S
(st response to changes in the recei).11 E -.165(ve)-.275 G 2.75(rs).165
G(et and/or netw)-2.75 E(ork conditions.)-.11 E(Rizzo/Iannaccone/V)72
769 Q(icisano/Handle)-.66 E 234.737(y[)-.165 G -.165(Pa)-234.737 G
(ge 1]).165 E EP
%%Page: 2 2
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E/F1 13/Times-Bold@0 SF -1.196
(Ta)239.126 85 S(ble of Contents)1.196 E/F2 10/Times-Roman@0 SF
(1. Introduction)97 123 Q F0 11(......................)3.56 G F2(3)11.5
E(1.1. T)107 135 Q(erminology)-.7 E F0 11(....................)11.48 G
F2(3)11.5 E(2. Protocol Ov)97 147 Q(ervie)-.15 E(w)-.25 E F0 11
(....................)5.08 G F2(3)11.5 E(2.1. P)107 159 Q(ack)-.15 E
(et Contents)-.1 E F0 11(...................)12.28 G F2(4)11.5 E
(2.1.1. Data P)117 171 Q(ack)-.15 E(ets)-.1 E F0 11(...................)
7.57 G F2(5)11.5 E(2.1.2. Feedback P)117 183 Q(ack)-.15 E(ets)-.1 E F0
11(..................)1.88 G F2(5)11.5 E(2.1.3. Field sizes and formats)
117 195 Q F0 11(................)8.01 G F2(6)11.5 E(2.2. W)107 207 Q
(indo)-.4 E(w-based controller)-.25 E F0 11(.................)2.98 G F2
(6)11.5 E(2.3. Ack)107 219 Q(er Selection)-.1 E F0 11
(...................)12.7 G F2(8)11.5 E(2.3.1. Initial Ack)117 231 Q
(er election)-.1 E F0 11(.................)1.87 G F2(8)11.5 E
(2.3.2. Ack)117 243 Q(er dropouts)-.1 E F0 11(..................)11.17 G
F2(9)11.5 E(2.4. TCP Throughput Equation)107 255 Q F0 11
(................)11.33 G F2(9)11.5 E(2.5. R)107 267 Q(TT measurement)
-.6 E F0 11(...................)2.64 G F2(10)6.5 E(2.5.1. Explicit T)117
279 Q(imestamp)-.35 E F0 11(.................)7.93 G F2(10)6.5 E
(2.5.2. Implicit timestamp)117 291 Q F0 11(.................)10.91 G F2
(10)6.5 E(2.5.3. Sequence numbers)117 303 Q F0 11(.................)
11.49 G F2(10)6.5 E(2.5.4. Recommendations)117 315 Q F0 11
(.................)12.31 G F2(11)6.5 E(2.6. Loss rate measurement)107
327 Q F0 11(.................)12.05 G F2(11)6.5 E(2.7. T)107 339 Q
(imeouts)-.35 E F0 11(.....................)11.82 G F2(12)6.5 E
(2.8. Interaction with feedback suppression schemes)107 351 Q F0 11
(..........)12.21 G F2(12)6.5 E(2.9. Interaction with ECN)107 363 Q F0
11(..................)4.4 G F2(12)6.5 E(3. Procedures - Sender)97 375 Q
F0 11(...................)13.72 G F2(13)6.5 E(4. Procedures -- Recei)97
387 Q -.15(ve)-.25 G(r).15 E F0 11(...................)3.02 G F2(14)6.5
E(5. Security Considerations)97 399 Q F0 11(..................)12.17 G
F2(15)6.5 E(6. Authors' Addresses)97 411 Q F0 11(....................)
1.35 G F2(15)6.5 E(7. Ackno)97 423 Q(wledgments)-.25 E F0 11
(....................)5.76 G F2(15)6.5 E(8. Full Cop)97 435 Q
(yright Statement)-.1 E F0 11(..................)6.42 G F2(16)6.5 E F0
(Rizzo/Iannaccone/V)72 769 Q(icisano/Handle)-.66 E 234.737(y[)-.165 G
-.165(Pa)-234.737 G(ge 2]).165 E EP
%%Page: 3 3
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E/F1 11/Times-Bold@0 SF(1.)72 85
Q/F2 14/Times-Bold@0 SF(Intr)5.5 E(oduction)-.252 E F0(This document de\
scribes PGMCC, a single rate multicast congestion control scheme which \
is)72 101.6 Q(TCP-friendly and achie)72 114.6 Q -.165(ve)-.275 G 2.75
(ss).165 G(calability)-2.75 E 2.75(,s)-.715 G(tability and f)-2.75 E
(ast response to v)-.11 E(ariations in netw)-.275 E(ork)-.11 E
(conditions.)72 127.6 Q(PGMCC is designed for multicast sessions with o\
ne sender and one or more recei)72 153.6 Q -.165(ve)-.275 G
(rs, and is a).165 E(good match for transport protocols using ne)72
166.6 Q -.055(ga)-.165 G(ti).055 E .33 -.165(ve a)-.275 H(ckno).165 E
(wledgements \(N)-.275 E(AKs\) to collect feedback)-.385 E
(from the recei)72 179.6 Q -.165(ve)-.275 G 2.75(rs. The).165 F
(congestion control scheme implemented by PGMCC closely mimics the)2.75
E(congestion-control beha)72 192.6 Q(viour of TCP)-.22 E 2.75(,a)-1.221
G 2.75(si)-2.75 G 2.75(tu)-2.75 G(ses a windo)-2.75 E
(w-based control loop which is run)-.275 E
(between the sender and a selected recei)72 205.6 Q -.165(ve)-.275 G
2.75(rc).165 G(alled the A)-2.75 E(CKER. The role of the A)-.44 E
(CKER is to)-.44 E(pro)72 218.6 Q(vide timely feedback in the same w)
-.165 E(ay as a TCP recei)-.11 E -.165(ve)-.275 G(r; additionally).165 E
2.75(,t)-.715 G(he A)-2.75 E(CKER is selected)-.44 E
(dynamically among the recei)72 231.6 Q -.165(ve)-.275 G
(rs as the one which w).165 E(ould e)-.11 E(xperience the lo)-.165 E
(west throughput if)-.275 E(separate TCP sessions were run between the \
sender and each of the recei)72 244.6 Q -.165(ve)-.275 G(rs.).165 E
(Scalability in PGMCC comes from the use of ne)72 270.6 Q -.055(ga)-.165
G(ti).055 E .33 -.165(ve a)-.275 H(ckno).165 E(wledgements \(N)-.275 E
(AKs\) for collecting)-.385 E(feedback from recei)72 283.6 Q -.165(ve)
-.275 G(rs other than the A).165 E 2.75(CKER. As)-.44 F 2.75(ac)2.75 G
(onsequence, the usual techniques for)-2.75 E -.385(NA)72 296.6 S 2.75
(Ks).385 G(uppression and aggre)-2.75 E -.055(ga)-.165 G
(tion can be used to reduce the amount of feedback to the source).055 E
(and impro)72 309.6 Q .33 -.165(ve t)-.165 H
(he scalability of the scheme.).165 E(PGMCC is designed to completely d\
ecouple congestion control from data inte)72 335.6 Q(grity)-.165 E 2.75
(.A)-.715 G 2.75(sa)-2.75 G(consequence, the scheme can w)72 348.6 Q
(ork with both reliable data transfer and unreliable communication)-.11
E(protocols such as those used for video or audio streaming.)72 361.6 Q
(While designed with multicast in mind, PGMCC can be equally used as a \
replacement for TCP)72 387.6 Q(for unicast sessions which require a lo)
72 400.6 Q(wer de)-.275 E(gree of reliability than what TCP of)-.165 E
(fers.)-.275 E F1(1.1.)72 439.6 Q/F3 13/Times-Bold@0 SF -1.196(Te)5.5 G
(rminology)1.196 E F0(In this document, the k)72 456.2 Q .33 -.165(ey w)
-.11 H(ords "MUST", "MUST NO).055 E(T", "REQ)-.44 E
(UIRED", "SHALL", "SHALL)-.11 E(NO)72 469.2 Q(T", "SHOULD", "SHOULD NO)
-.44 E(T", "RECOMMENDED", "MA)-.44 E(Y", and "OPTION)-1.155 E
(AL" are to)-.385 E
(be interpreted as described in RFC 2119 and indicate requirement le)72
482.2 Q -.165(ve)-.275 G(ls for compliant PGMCC).165 E(implementations.)
72 495.2 Q F1(2.)72 534.2 Q F2(Pr)5.5 E(otocol Ov)-.252 E(er)-.14 E
(view)-.14 E F0(PGMCC is based on tw)72 550.8 Q 2.75(os)-.11 G
(eparate b)-2.75 E(ut complementary mechanisms:)-.22 E 11(oA)77.5 567.4
S(windo)-8.25 E
(w-based control loop which closely emulates TCP congestion control.)
-.275 E(The windo)94 580.4 Q(w-based control loop is simply an adaptati\
on of the TCP congestion control)-.275 E
(scheme to transport protocols where missing \(because of netw)94 593.4
Q(ork errors or congestion\) data)-.11 E(pack)94 606.4 Q(ets are not ne\
cessarily retransmitted, and so the congestion control scheme cannot re\
ly)-.11 E(on cumulati)94 619.4 Q .33 -.165(ve a)-.275 H(ckno).165 E 2.75
(wledgements. In)-.275 F(PGMCC, the `)2.75 E(`congestion windo)-.814 E
(w')-.275 E 2.75('i)-.814 G 2.75(ss)-2.75 G(imulated)-2.75 E
(using a tok)94 632.4 Q
(en-based scheme which permits congestion control to be decoupled from)
-.11 E(retransmission state. One of the recei)94 645.4 Q -.165(ve)-.275
G(rs in the group operates as the A).165 E(CKER, i.e. the node)-.44 E
(in char)94 658.4 Q(ge of sending positi)-.198 E .33 -.165(ve a)-.275 H
(ckno).165 E(wledgements back to the source and thus controlling the)
-.275 E(rate of the transfer)94 671.4 Q(.)-.605 E 11(oA)77.5 701 S
(procedure to select the A)-8.25 E(CKER.)-.44 E
(The purpose of this procedure is to mak)94 714 Q 2.75(es)-.11 G
(ure that, in presence of multiple recei)-2.75 E -.165(ve)-.275 G
(rs, the).165 E(Rizzo/Iannaccone/V)72 769 Q(icisano/Handle)-.66 E
185.237(yS)-.165 G(ection 2.)-185.237 E([P)5.5 E(age 3])-.165 E EP
%%Page: 4 4
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E -.44(AC)94 85 S
(KER is dynamically selected to be the recei).44 E -.165(ve)-.275 G 2.75
(rw).165 G(hich w)-2.75 E(ould ha)-.11 E .33 -.165(ve t)-.22 H(he lo)
.165 E(west throughput)-.275 E
(if separate TCP sessions were run between the sender and each recei)94
98 Q -.165(ve)-.275 G -.605(r.).165 G -.165(Fo)94 111 S 2.75(rt).165 G
(he ack)-2.75 E(er selection mechanism, PGMCC uses a throughput equatio\
n to determine the)-.11 E -.165(ex)94 124 S(pected throughput for a gi)
.165 E -.165(ve)-.275 G 2.75(nr).165 G(ecei)-2.75 E -.165(ve)-.275 G
2.75(ra).165 G 2.75(saf)-2.75 G
(unction of the loss rate and round-trip time.)-2.75 E(Unlik)94 137 Q
2.75(eo)-.11 G(ther schemes [2], the TCP throughput equation is not use\
d to determine the actual)-2.75 E
(sending rate, which is completely controlled by the windo)94 150 Q
(w-based control loop.)-.275 E(In principle, PGMCC')72 179.6 Q 2.75(sc)
-.605 G(ongestion control mechanism w)-2.75 E(orks as follo)-.11 E(ws:)
-.275 E 11(oR)77.5 209.2 S(ecei)-11 E -.165(ve)-.275 G
(rs measure the loss rate and feed this information back to the sender)
.165 E 2.75(,e)-.44 G(ither in A)-2.75 E(CK)-.44 E(or N)94 222.2 Q
(AK messages.)-.385 E 11(oT)77.5 251.8 S(he sender also uses these feed\
back messages to measure the round-trip time \(R)-11 E(TT\) to each)-.66
E(recei)94 264.8 Q -.165(ve)-.275 G -.605(r.).165 G 11(oT)77.5 294.4 S
(he loss rate and R)-11 E(TT are then fed into PGMCC')-.66 E 2.75(st)
-.605 G(hroughput equation, to determine the)-2.75 E -.165(ex)94 307.4 S
(pected throughput to that recei).165 E -.165(ve)-.275 G -.605(r.).165 G
11(oT)77.5 337 S(he sender then selects as the ack)-11 E(er the recei)
-.11 E -.165(ve)-.275 G 2.75(rw).165 G(ith the lo)-2.75 E(west e)-.275 E
(xpected throughput, as)-.165 E(computed by the equation.)94 350 Q
(The dynamics of the ack)72 366.6 Q(er selection mechanism are sensiti)
-.11 E .33 -.165(ve t)-.275 H 2.75(oh).165 G .55 -.275(ow t)-2.75 H
(he measurements are).275 E(performed and applied. In the rest of this \
document we suggest speci\214c mechanisms to perform)72 379.6 Q
(and apply these measurements.)72 392.6 Q
(Other mechanisms are possible, b)5.5 E
(ut it is important to understand)-.22 E(ho)72 405.6 Q 2.75(wt)-.275 G
(he interactions between mechanisms af)-2.75 E
(fect the dynamics of PGMCC.)-.275 E/F1 11/Times-Bold@0 SF(2.1.)72 444.6
Q/F2 13/Times-Bold@0 SF -.13(Pa)5.5 G(ck).13 E(et Contents)-.13 E F0
(Before specifying the sender and recei)72 461.2 Q -.165(ve)-.275 G 2.75
(rf).165 G(unctionality)-2.75 E 2.75(,w)-.715 G 2.75(ed)-2.75 G
(escribe the information required by)-2.75 E(PGMCC to perform its tasks\
. This information is carried in the data pack)72 474.2 Q
(ets sent by the sender)-.11 E(,)-.44 E(and in the feedback pack)72
487.2 Q(ets sent by the recei)-.11 E -.165(ve)-.275 G 3.96 -.605(r. A)
.165 H 2.75(sP).605 G(GMCC will be used along with some)-2.75 E
(transport protocol, the actual data and feedback pack)72 500.2 Q
(ets will contain further information for use)-.11 E
(by the protocol itself. F)72 513.2 Q
(or this reason, we do not specify pack)-.165 E
(et formats, as these depend on the)-.11 E
(details of the transport protocol used.)72 526.2 Q
(Note that the requirements of the transport protocol in terms of pack)
72 552.2 Q(et generation may dif)-.11 E(fer from)-.275 E
(those of PGMCC. As an e)72 565.2 Q(xample, most N)-.165 E
(AK-based reliable multicast protocols do not use)-.385 E(positi)72
578.2 Q .33 -.165(ve a)-.275 H(ckno).165 E(wledgements, b)-.275 E
(ut PGMCC requires A)-.22 E(CKs for clocking out data pack)-.44 E
(ets; unreliable)-.11 E(transport protocols might ha)72 591.2 Q .33
-.165(ve n)-.22 H 2.75(oi).165 G(nterest in generating N)-2.75 E
(AKs for data inte)-.385 E(grity purposes, yet)-.165 E
(PGMCC depends on N)72 604.2 Q
(AKs reaching the data sender in order to elect the A)-.385 E(CKER.)-.44
E(Implementors may decide to insert PGMCC-related information in alread\
y e)72 633.8 Q(xisting protocol)-.165 E(pack)72 646.8 Q(ets whene)-.11 E
-.165(ve)-.275 G 2.75(rp).165 G(ossible, b)-2.75 E
(ut in cases such as the ones described in the pre)-.22 E
(vious paragraph, it)-.275 E
(might be necessary to de\214ne and generate ne)72 659.8 Q 2.75(wp)-.275
G(ack)-2.75 E(ets e)-.11 E(xclusi)-.165 E -.165(ve)-.275 G
(ly for congestion control).165 E(purposes. As an e)72 672.8 Q(xample, \
in a prototype implementation of PGMCC on top of the PGM protocol)-.165
E([7], some of the information used by PGMCC is already present in the \
original protocol pack)72 685.8 Q(ets,)-.11 E
(and PGMCC-speci\214c information is carried as PGM options in OD)72
698.8 Q -1.331 -1.221(AT A)-.44 H(and N)3.971 E(AK pack)-.385 E(ets.)
-.11 E(Ho)72 711.8 Q(we)-.275 E -.165(ve)-.275 G .88 -.44(r, a n).165 H
.55 -.275(ew p).44 H(ack).275 E(et type has been de\214ned for A)-.11 E
(CKs, which are generated according to the)-.44 E
(rules de\214ned in this document.)72 724.8 Q(Rizzo/Iannaccone/V)72 769
Q(icisano/Handle)-.66 E 176.987(yS)-.165 G(ection 2.1.)-176.987 E([P)5.5
E(age 4])-.165 E EP
%%Page: 5 5
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E/F1 11/Times-Bold@0 SF 2.75
(2.1.1. Data)72 85 R -.11(Pa)2.75 G(ck).11 E(ets)-.11 E F0
(Each data pack)72 101.6 Q
(et sent by the data sender contains the follo)-.11 E(wing information:)
-.275 E 11(oA)77.5 131.2 S(SEQ)-8.25 E
(UENCE NUMBER. This number is incremented by one for each data pack)-.11
E(et)-.11 E 2.75(transmitted. The)94 144.2 R(\214eld must be suf)2.75 E
(\214ciently lar)-.275 E(ge that it does not wrap causing tw)-.198 E
2.75(od)-.11 G(if)-2.75 E(ferent)-.275 E(pack)94 157.2 Q
(ets with the same sequence number to be in the recei)-.11 E -.165(ve)
-.275 G(r').165 E 2.75(sr)-.605 G(ecent pack)-2.75 E(et history at the)
-.11 E(same time.)94 170.2 Q 11(oA)77.5 199.8 S(TIMEST)-8.25 E
(AMP \(or equi)-1.023 E -.275(va)-.275 G
(lent information, see Section 2.5\) indicating when the pack).275 E(et)
-.11 E(with this sequence number has been sent.)94 212.8 Q
(There is no requirement for synchronized clocks)5.5 E
(between the sender and the recei)94 225.8 Q -.165(ve)-.275 G 2.75
(rs. The).165 F(timestamp is used to measure netw)2.75 E(ork round-trip)
-.11 E(times, so needs suf)94 238.8 Q
(\214cient resolution for this task.)-.275 E 2.75(Ar)5.5 G
(esolution of 1ms w)-2.75 E(ould be adequate.)-.11 E 11(oT)77.5 268.4 S
(he A)-11 E(CKER IDENTITY)-.44 E 2.75(,i)-1.419 G
(.e. the identity of the recei)-2.75 E -.165(ve)-.275 G 2.75(ri).165 G
2.75(nc)-2.75 G(har)-2.75 E(ge of sending an)-.198 E(ackno)94 281.4 Q
(wledgement for this data pack)-.275 E 2.75(et. The)-.11 F -.44(AC)2.75
G(KER is elected as a result of the process).44 E
(described in Section 2.3.)94 294.4 Q 2.75(As)94 307.4 S(pecial v)-2.75
E(alue is used to indicate that no A)-.275 E
(CKER is designated for this pack)-.44 E(et -- this can)-.11 E
(happen at the be)94 320.4 Q(ginning of a session or when the current A)
-.165 E(CKER lea)-.44 E -.165(ve)-.22 G 2.75(st).165 G(he group.)-2.75 E
(Recei)94 333.4 Q -.165(ve)-.275 G(rs interpret this v).165 E
(alue as a request to elect a ne)-.275 E 2.75(wa)-.275 G(ck)-2.75 E(er)
-.11 E(.)-.605 E F1 2.75(2.1.2. F)72 372.4 R(eedback P)-.275 E(ack)-.11
E(ets)-.11 E F0(There are tw)72 389 Q 2.75(ot)-.11 G
(ypes of feedback pack)-2.75 E(ets used by PGMCC: A)-.11 E(CK pack)-.44
E(ets and N)-.11 E(AK pack)-.385 E(ets.)-.11 E -.44(AC)72 402 S 2.75(Kp)
.44 G(ack)-2.75 E(ets are generated by the current A)-.11 E
(CKER, and are used to detect loss or successful)-.44 E(deli)72 415 Q
-.165(ve)-.275 G(ry of pack).165 E(ets, and to re)-.11 E
(gulate the throughput accordingly)-.165 E 2.75(.A)-.715 G(CK pack)-3.19
E(ets also contain)-.11 E(information used to determine the TCP-equi)72
428 Q -.275(va)-.275 G(lent throughput for the A).275 E(CKER.)-.44 E
-.385(NA)72 441 S 2.75(Kp).385 G(ack)-2.75 E(ets are sent by an)-.11 E
2.75(yr)-.165 G(ecei)-2.75 E -.165(ve)-.275 G 2.75(rw).165 G(ho e)-2.75
E(xperiences loss.)-.165 E(The)5.5 E 2.75(yc)-.165 G
(ontain information used to)-2.75 E(determine the TCP-equi)72 454 Q
-.275(va)-.275 G(lent throughput for that recei).275 E -.165(ve)-.275 G
1.21 -.605(r. I).165 H 2.75(na).605 G 2.75(na)-2.75 G
(ctual protocol instantiation)-2.75 E(\(such as PGM [7]\), N)72 467 Q
(AK pack)-.385 E
(ets might also be used by the protocol to request the retransmission)
-.11 E(of speci\214c pack)72 480 Q
(ets, and indicate the identity of the pack)-.11 E(et being requested.)
-.11 E(Both A)72 506 Q(CK and N)-.44 E(AK pack)-.385 E
(ets are sent by data recei)-.11 E -.165(ve)-.275 G
(rs, and contain the follo).165 E(wing information:)-.275 E 11(oT)77.5
535.6 S(he TIMEST)-11 E(AMP \(or equi)-1.023 E -.275(va)-.275 G
(lent information\) deri).275 E -.165(ve)-.275 G 2.75(df).165 G
(rom the most recently recei)-2.75 E -.165(ve)-.275 G 2.75(dd).165 G
(ata)-2.75 E(pack)94 548.6 Q
(et according to one of the techniques described in Section 2.5.)-.11 E
(This v)94 561.6 Q(alue is used by the sender to measure the R)-.275 E
(TT to the recei)-.66 E -.165(ve)-.275 G 2.75(rw).165 G
(ho generated this)-2.75 E(feedback pack)94 574.6 Q(et.)-.11 E 11(o`)
77.5 604.2 S(`p')-11.814 E(', the recei)-.814 E -.165(ve)-.275 G(r').165
E 2.75(sc)-.605 G(urrent estimate of the LOSS RA)-2.75 E 2.75(TE. The)
-1.221 F(loss rate is measured by)2.75 E(recei)94 617.2 Q -.165(ve)-.275
G(rs as described in Section 2.6).165 E(In addition to the abo)72 633.8
Q -.165(ve)-.165 G 2.75(,A).165 G(CK pack)-3.19 E(ets \(sent by the ack)
-.11 E(er designated in the corresponding data)-.11 E(pack)72 646.8 Q
(ets\) must also contain the follo)-.11 E(wing information:)-.275 E 11
(oR)77.5 663.4 S(X_MAX, the highest sequence number among recei)-11 E
-.165(ve)-.275 G 2.75(dd).165 G(ata pack)-2.75 E
(ets \(taking care to deal)-.11 E
(with sequence number wrapping correctly\).)94 676.4 Q 11(oA)77.5 693 S
(CK_BITMAP)-11.44 E 2.75(,ab)-1.221 G(itmap indicating the recei)-2.75 E
.33 -.165(ve s)-.275 H(tatus of the latest N \(typically N=32\) data)
.165 E(pack)94 706 Q
(ets with sequence numbers RX_MAX-\(N-1\) to RX_MAX.)-.11 E
(Rizzo/Iannaccone/V)72 769 Q(icisano/Handle)-.66 E 168.737(yS)-.165 G
(ection 2.1.2.)-168.737 E([P)5.5 E(age 5])-.165 E EP
%%Page: 6 6
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E
(This information is used by the sender to record which pack)72 85 Q
(ets ha)-.11 E .33 -.165(ve b)-.22 H(een recei).165 E -.165(ve)-.275 G
2.75(do).165 G 2.75(rl)-2.75 G(ost, and)-2.75 E
(manipulate the transmit windo)72 98 Q 2.75(wa)-.275 G(ccordingly)-2.75
E 5.5(.N)-.715 G(ote that each A)-5.5 E(CK pack)-.44 E
(et contains information)-.11 E(about multiple pack)72 111 Q
(ets, and this increases the rob)-.11 E
(ustness of the scheme to loss of A)-.22 E(CK pack)-.44 E(ets.)-.11 E
(This is necessary because A)72 124 Q(CKs are not sent reliably \(unlik)
-.44 E 2.75(eT)-.11 G(CP')-2.75 E 2.75(sA)-.605 G(CKs, which are)-3.19 E
(cumulati)72 137 Q -.165(ve)-.275 G(\).).165 E/F1 11/Times-Bold@0 SF
2.75(2.1.3. Field)72 176 R(sizes and f)2.75 E(ormats)-.275 E F0
(The follo)72 192.6 Q(wing sizes and formats are suggested for the v)
-.275 E(arious \214elds used by PGMCC and)-.275 E(transmitted o)72 205.6
Q -.165(ve)-.165 G 2.75(rt).165 G(he netw)-2.75 E(ork:)-.11 E 11(oS)77.5
235.2 S(EQ)-11 E(UENCE NUMBERS)-.11 E(32 bit, unsigned, netw)94 248.2 Q
(ork order)-.11 E(.)-.605 E 11(oT)77.5 277.8 S(IMEST)-11 E(AMPS)-1.023 E
(32 bit, unsigned, netw)94 290.8 Q(ork order)-.11 E 5.5(.A)-.605 G
(resolution of 1ms or better is desirable.)-2.75 E 11(oA)77.5 320.4 S
(CKER IDENTITY)-11.44 E(Same size and format of a netw)94 333.4 Q
(ork layer address \(e.g. 32 bit for IPv4\).)-.11 E(Note though that)5.5
E(using an IP address for the Ack)94 346.4 Q
(er Identify will cause problems with N)-.11 E 2.442 -1.221(AT t)-.385 H
(ra)1.221 E -.165(ve)-.22 G(rsal.).165 E -.385(Tr)94 359.4 S
(ansport protocol designers might e).385 E
(xamine the SSRC mechanism used by R)-.165 E(TP [6] as an)-.66 E
(alternati)94 372.4 Q .33 -.165(ve f)-.275 H
(orm of node identi\214er that could be used as Ack).165 E(er Identity)
-.11 E(.)-.715 E 11(oL)77.5 402 S(OSS RA)-11 E(TE \(`)-1.221 E(`p')-.814
E('\))-.814 E(16-bit unsigned inte)94 415 Q(ger)-.165 E 2.75(,i)-.44 G
2.75(nn)-2.75 G(etw)-2.75 E
(ork format, with 0 indicating no loss and 2^16-1 indicating)-.11 E
(100% loss.)94 428 Q 11(oA)77.5 457.6 S(CK BITMAP)-11.44 E
(32-bit, in netw)94 470.6 Q
(ork format, with least signi\214cant bit indicating recei)-.11 E .33
-.165(ve s)-.275 H(tatus of pack).165 E(et)-.11 E(RX_MAX.)94 483.6 Q F1
(2.2.)72 522.6 Q/F2 13/Times-Bold@0 SF -.234(Wi)5.5 G(ndo).234 E
(w-based contr)-.13 E(oller)-.234 E F0(In a windo)72 539.2 Q
(w-based congestion control scheme such as TCP)-.275 E 2.75(,t)-1.221 G
(he `)-2.75 E(`congestion windo)-.814 E(w')-.275 E 2.75('r)-.814 G
(epresents,)-2.75 E(among other things, the maximum amount of pack)72
552.2 Q(ets in \215ight at an)-.11 E 2.75(yt)-.165 G
(ime, which in turn controls)-2.75 E
(the throughput of the session. The sender k)72 565.2 Q
(eeps track of the actual number of pack)-.11 E(ets in \215ight,)-.11 E
(basing on its transmissions and the reception of ackno)72 578.2 Q
(wledgements.)-.275 E
(The sender may dynamically change the size of the windo)72 604.2 Q 1.43
-.715(w, a)-.275 H(ccording to the congestion control).715 E
(scheme being used. In TCP)72 617.2 Q 2.75(,a)-1.221 G(nd PGMCC, an `)
-2.75 E -.88(`A)-.814 G(dditi).88 E .33 -.165(ve I)-.275 H
(ncrease Multiplicati).165 E .33 -.165(ve D)-.275 H(ecrease').165 E(')
-.814 E(\(AIMD\) scheme is used: in absence of loss, the windo)72 630.2
Q 2.75(wi)-.275 G 2.75(si)-2.75 G(ncreased by some \214x)-2.75 E
(ed amount)-.165 E(\(typically one pack)72 643.2 Q
(et\) per round trip time \(R)-.11 E(TT\), whereas upon loss the windo)
-.66 E 2.75(wi)-.275 G 2.75(sr)-2.75 G(educed to a)-2.75 E
(fraction of its original v)72 656.2 Q(alue \(typically halv)-.275 E
(ed\) in each R)-.165 E(TT in which a loss e)-.66 E -.165(ve)-.275 G
(nt is e).165 E(xperienced.)-.165 E(In PGMCC the windo)72 682.2 Q 2.75
(wi)-.275 G 2.75(sm)-2.75 G(anaged using a tok)-2.75 E
(en-based mechanism, controlled by tw)-.11 E 2.75(ov)-.11 G(ariables:)
-3.025 E 11(oA)77.5 698.8 S -.814(``)-8.25 G -.44(Wi).814 G(ndo).44 E
2.75(wS)-.275 G(ize')-2.75 E(', W)-.814 E 2.75(,w)-1.012 G
(hich describes the current windo)-2.75 E 2.75(ws)-.275 G(ize in pack)
-2.75 E(ets.)-.11 E(Rizzo/Iannaccone/V)72 769 Q(icisano/Handle)-.66 E
176.987(yS)-.165 G(ection 2.2.)-176.987 E([P)5.5 E(age 6])-.165 E EP
%%Page: 7 7
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E 11(oA)77.5 85 S -.814(``)-8.25
G -.88(To).814 G -.11(ke).88 G 2.75(nC).11 G(ount')-2.75 E(', T)-.814 E
2.75(,w)-.814 G(hich indicates the number of pack)-2.75 E
(ets that can be transmitted.)-.11 E 2.75(Ti)5.5 G(s)-2.75 E
(bounded abo)94 98 Q .33 -.165(ve b)-.165 H 2.75(yW).165 G 5.5(.I)-3.762
G 2.75(ti)-5.5 G 2.75(sd)-2.75 G(ecremented e)-2.75 E -.165(ve)-.275 G
(ry time a pack).165 E(et is transmitted, and incremented)-.11 E -2.365
-.275(ev e)94 111 T(ry time an A).275 E(CK is recei)-.44 E -.165(ve)
-.275 G(d, according to the rules belo).165 E -.715(w.)-.275 G
(Note that these tw)72 127.6 Q 2.75(ov)-.11 G
(ariables need to hold non-inte)-3.025 E(ger data.)-.165 E -.88(Ty)5.5 G
(pically a \214x).88 E(ed point representation)-.165 E
(with at least 16 bits for both inte)72 140.6 Q
(ger and fractional parts w)-.165 E
(ould be acceptable for implementation)-.11 E(purposes.)72 153.6 Q
(The information contained in each A)72 170.2 Q
(CK is used to determine ho)-.44 E 2.75(wm)-.275 G(an)-2.75 E 2.75(yn)
-.165 G .55 -.275(ew p)-2.75 H(ack).275 E(ets are)-.11 E(ackno)72 183.2
Q(wledged by that A)-.275 E(CK, and whether there are unackno)-.44 E
(wledged pack)-.275 E(ets which were not)-.11 E(reported in pre)72 196.2
Q(vious A)-.275 E 2.75(CKs. The)-.44 F
(sender also schedules a timeout to react in case no A)2.75 E(CKs are)
-.44 E(recei)72 209.2 Q -.165(ve)-.275 G(d.).165 E(The sender beha)72
235.2 Q -.165(ve)-.22 G 2.75(sa).165 G 2.75(sf)-2.75 G(ollo)-2.75 E(ws:)
-.275 E 11(oI)77.5 264.8 S(NITIALIZA)-11 E(TION)-1.221 E
(At startup, or after a timeout, both W and T are set to 1.)94 277.8 Q
11(oA)77.5 307.4 S(CK RECEPTION, NO LOSS DETECTED)-11.44 E
(If the incoming A)94 320.4 Q(CK reports ne)-.44 E 2.75(wa)-.275 G(ckno)
-2.75 E(wledged pack)-.275 E(ets, and no loss \(as de\214ned in the ne)
-.11 E(xt)-.165 E(paragraph\) is detected, then the windo)94 333.4 Q
2.75(wi)-.275 G 2.75(si)-2.75 G(n\215ated by one pack)-2.75 E(et per R)
-.11 E(TT)-.66 E(.)-.814 E(NO)94 359.4 Q(TE: during the slo)-.44 E
(w-start phase, TCP opens the windo)-.275 E 2.75(we)-.275 G
(xponentially up to the)-2.915 E(SSTHRESH v)94 372.4 Q(alue, which is c\
omputed by TCP according to the dynamics of the session and)-.275 E
(updated upon losses.)94 385.4 Q 1.76 -.88(We d)94 411.4 T 2.75(or).88 G
(ecommend that PGMCC uses a similar strate)-2.75 E(gy)-.165 E 2.75(,b)
-.715 G(ut using a \214x)-2.97 E(ed, small v)-.165 E(alue for)-.275 E
(SSTHRESH \(e.g. 4 pack)94 424.4 Q 2.75(ets\). In)-.11 F -.11(fa)2.75 G
(ct, due to the dynamicity of the A).11 E(CKER, which might)-.44 E
(change on e)94 437.4 Q -.165(ve)-.275 G(ry single pack).165 E
(et, it is hard to compute a reliable estimate of the SSTHRESH)-.11 E
(without k)94 450.4 Q(eeping state for multiple recei)-.11 E -.165(ve)
-.275 G(rs, and the bene\214ts are small in an).165 E 2.75(ye)-.165 G
-.165(ve)-3.025 G(nt.).165 E(In summary)94 476.4 Q 2.75(,t)-.715 G
(he reaction to A)-2.75 E
(CK reception on no loss modi\214es T and W as follo)-.44 E
(ws \(here, N)-.275 E(is the number of ne)94 489.4 Q 2.75(wp)-.275 G
(ack)-2.75 E(ets ackno)-.11 E(wledged by the incoming A)-.275 E(CK\):)
-.44 E(if \(W < SSTHRESH\) then)99.5 515.4 Q 2.75(D=m)110.5 528.4 S
(in\(N, SSTHRESH - W\) // use the \214rst D acks for e)-2.75 E
(xp.opening)-.165 E 54.307 2.75(N=N-D /)110.5 541.4 T 2.75(/a)-2.75 G
(nd the remaining ones for linear opening)-2.75 E 2.75(T=T+2)110.5 554.4
S(*D)-2.75 E 2.75(W=W+D)110.5 567.4 S(endif)99.5 580.4 Q
(// do linear windo)99.5 593.4 Q 2.75(wo)-.275 G
(pening with the remaining acks)-2.75 E 2.75(T=T+N*\(1+1)99.5 606.4 S
(/W \))-2.75 E 2.75(W=W+N)99.5 619.4 S(/W)-2.75 E 11(oP)77.5 649 S -.44
(AC)-12.012 G(KET TRANSMISSION).44 E(One tok)94 662 Q
(en is consumed for each pack)-.11 E(et transmitted:)-.11 E 2.75(T=T-1)
279.095 688 S(Rizzo/Iannaccone/V)72 769 Q(icisano/Handle)-.66 E 176.987
(yS)-.165 G(ection 2.2.)-176.987 E([P)5.5 E(age 7])-.165 E EP
%%Page: 8 8
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E 11(oA)77.5 85 S
(CK RECEPTION, LOSS DETECTED)-11.44 E(If the incoming A)94 98 Q
(CK reports an unackno)-.44 E(wledged data pack)-.275 E
(et which is follo)-.11 E(wed by at least 3)-.275 E(ackno)94 111 Q
(wledged data pack)-.275 E(ets, then the pack)-.11 E
(et is assumed to be lost and PGMCC reacts by)-.11 E(halving the windo)
94 124 Q 1.43 -.715(w, i)-.275 H 2.75(nt).715 G(he same w)-2.75 E
(ay as TCP after 3 duplicate ackno)-.11 E 2.75(wledgements. This)-.275 F
(is)2.75 E(achie)94 137 Q -.165(ve)-.275 G 2.75(db).165 G 2.75(ym)-2.75
G(odifying T and W as follo)-2.75 E(ws:)-.275 E 2.75(T=T-W)247.734 163 S
(/2 , W = W/2)-2.75 E(to simulate the multiplicati)94 189 Q .33 -.165
(ve d)-.275 H(ecrease.).165 E(Additionally)94 202 Q 2.75(,a)-.715 G
(ll windo)-2.75 E 2.75(wm)-.275 G
(anipulation is suspended for the subsequent R)-2.75 E(TT)-.66 E 5.5(.T)
-.814 G(his is)-5.5 E(achie)94 215 Q -.165(ve)-.275 G 2.75(db).165 G
2.75(yr)-2.75 G(ecording the current transmit sequence number)-2.75 E
2.75(,a)-.44 G(nd canceling an)-2.75 E 2.75(yf)-.165 G(urther)-2.75 E
(manipulation of the windo)94 228 Q 2.75(wu)-.275 G
(ntil feedback is recei)-2.75 E -.165(ve)-.275 G 2.75(df).165 G
(or the ne)-2.75 E(xt transmitted pack)-.165 E(et, or)-.11 E
(until a timeout occurs.)94 241 Q/F1 11/Times-Bold@0 SF(2.3.)72 296.6 Q
/F2 13/Times-Bold@0 SF(Ack)5.5 E(er Selection)-.13 E F0(The A)72 313.2 Q
(CKER selection process in PGMCC aims at locating the recei)-.44 E -.165
(ve)-.275 G 2.75(rw).165 G(hich w)-2.75 E(ould ha)-.11 E .33 -.165(ve t)
-.22 H(he).165 E(lo)72 326.2 Q(west throughput if each recei)-.275 E
-.165(ve)-.275 G 2.75(rw).165 G
(ere using a separate TCP connection to transfer data.)-2.75 E(Because \
the steady-state throughput of a TCP connection can be characterised in\
 a reasonably)72 352.2 Q(accurate w)72 365.2 Q(ay in terms of its loss \
rate and round trip time [3], the throughput for each recei)-.11 E -.165
(ve)-.275 G 2.75(rc).165 G(an)-2.75 E(be estimated by using these tw)72
378.2 Q 2.75(op)-.11 G(arameters.)-2.75 E(Whene)72 404.2 Q -.165(ve)
-.275 G 2.75(ra).165 G 2.75(nA)-2.75 G(CK or N)-3.19 E(AK pack)-.385 E
(et from an)-.11 E 2.75(yo)-.165 G 2.75(ft)-2.75 G(he recei)-2.75 E
-.165(ve)-.275 G(rs reaches it, the sender is able to).165 E
(compute the e)72 417.2 Q(xpected throughput T_i for that recei)-.165 E
-.165(ve)-.275 G 2.75(rb).165 G 2.75(yu)-2.75 G(sing the equation sho)
-2.75 E(wn in Section)-.275 E(2.4, with the round trip time R)72 430.2 Q
(TT and loss rate p and measured as described in Sections 2.5 and)-.66 E
(2.6, respecti)72 443.2 Q -.165(ve)-.275 G(ly).165 E 5.5(.A)-.715 G 2.75
(ta)-5.5 G .33 -.165(ny g)-2.75 H -2.365 -.275(iv e).165 H 2.75(nt).275
G(ime, the sender stores the e)-2.75 E
(xpected throughput for the current)-.165 E -.44(AC)72 456.2 S
(KER, T_ack).44 E(er)-.11 E 2.75(.T)-.605 G(his v)-2.75 E
(alue is updated e)-.275 E -.165(ve)-.275 G(ry time an A).165 E(CK or N)
-.44 E(AK from the current A)-.385 E(CKER is)-.44 E(recei)72 469.2 Q
-.165(ve)-.275 G 2.75(d\().165 G(note that, after a ne)-2.75 E 2.75(wA)
-.275 G(CKER is selected, the sender will typically recei)-3.19 E .33
-.165(ve A)-.275 H(CKs from)-.275 E(the old A)72 482.2 Q(CKER for one R)
-.44 E(TT)-.66 E 2.75(,a)-.814 G(nd the feedback from dif)-2.75 E
(ferent A)-.275 E(CKERs might be interlea)-.44 E -.165(ve)-.22 G 2.75
(di).165 G(f)-2.75 E(the paths leading to them ha)72 495.2 Q .33 -.165
(ve d)-.22 H(if).165 E(ferent round trip times\).)-.275 E(Whene)72 521.2
Q -.165(ve)-.275 G 2.75(ra).165 G 2.75(nA)-2.75 G(CK or N)-3.19 E
(AK is recei)-.385 E -.165(ve)-.275 G 2.75(df).165 G
(rom another node i \(a pre)-2.75 E(vious A)-.275 E(CKER or some other)
-.44 E(recei)72 534.2 Q -.165(ve)-.275 G(r\), the e).165 E(xpected thro\
ughput T_i for that node is computed, and compared with T_ack)-.165 E
(er)-.11 E(.)-.605 E(Node i is selected as the ne)72 547.2 Q 2.75(wa)
-.275 G(ck)-2.75 E(er if)-.11 E(T_i < C * T_ack)247.377 573.2 Q(er)-.11
E(where the constant C between 0 and 1 pro)72 599.2 Q(vides some h)-.165
E(ysteresis and a)-.055 E -.22(vo)-.22 G(ids too frequent).22 E
(oscillations in the choice of the A)72 612.2 Q 2.75(CKER. A)-.44 F
(suggested v)2.75 E(alue for C is 0.75.)-.275 E
(Note that, from an implementation point of vie)72 638.2 Q 2.75(w\()
-.275 G(see Section 2.4\), it is more con)-2.75 E -.165(ve)-.44 G
(nient to).165 E(compute T_i ^\(-2\), so the abo)72 651.2 Q .33 -.165
(ve e)-.165 H(quation must be modi\214ed accordingly).165 E(.)-.715 E F1
2.75(2.3.1. Initial)72 690.2 R(Ack)2.75 E(er election)-.11 E F0
(Upon reception of a data pack)72 706.8 Q(et reporting that no ack)-.11
E(er is currently selected, recei)-.11 E -.165(ve)-.275 G(rs generate a)
.165 E(dummy N)72 719.8 Q
(AK report which is used to elect the initial ack)-.385 E(er)-.11 E 2.75
(.T)-.605 G(he N)-2.75 E(AK is sent with the usual)-.385 E
(Rizzo/Iannaccone/V)72 769 Q(icisano/Handle)-.66 E 168.737(yS)-.165 G
(ection 2.3.1.)-168.737 E([P)5.5 E(age 8])-.165 E EP
%%Page: 9 9
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E(feedback suppression mechanis\
m dictated by the transport protocol \(possibly with shorter time)72 85
Q(constants\) to a)72 98 Q -.22(vo)-.22 G(id feedback implosion, and th\
e sender will select the source of the \214rst incoming).22 E -.385(NA)
72 111 S 2.75(Ka).385 G 2.75(st)-2.75 G(he ne)-2.75 E 2.75(wA)-.275 G
(CKER.)-3.19 E/F1 11/Times-Bold@0 SF 2.75(2.3.2. Ack)72 150 R(er dr)-.11
E(opouts)-.198 E F0(If the A)72 179.6 Q(CKER decides to disconnect from\
 the session, it can cause the session to stop. T)-.44 E 2.75(oa)-.88 G
-.22(vo)-2.97 G(id).22 E(this, it is recommended that an A)72 192.6 Q
(CKER deciding to lea)-.44 E .33 -.165(ve t)-.22 H
(he session informs the sender by).165 E(sending an A)72 205.6 Q
(CK pack)-.44 E(et \(or a duplicate\) carring an "A)-.11 E(CKER_LEA)-.44
E(VING" option.)-1.485 E(The reception)5.5 E(of this pack)72 218.6 Q
(et by the sender will in turn trigger an initial ack)-.11 E
(er election phase.)-.11 E F1(2.4.)72 270.6 Q/F2 13/Times-Bold@0 SF
(TCP Thr)5.5 E(oughput Equation)-.234 E F0(An)72 287.2 Q 2.75(yr)-.165 G
(ealistic equation of TCP throughput as a function of loss e)-2.75 E
-.165(ve)-.275 G(nt rate and R).165 E(TT should be)-.66 E
(suitable for use in PGMCC.)72 300.2 Q(Unlik)5.5 E 2.75(eo)-.11 G
(ther schemes [2] where the throughput equation directly)-2.75 E
(controls the transmit rate, in PGMCC the equation is used only for ack)
72 313.2 Q(er selection purposes, and)-.11 E(the throughput v)72 326.2 Q
(alues are only compared among themselv)-.275 E 2.75(es. As)-.165 F 2.75
(ac)2.75 G(onsequence, we can use the)-2.75 E(follo)72 339.2 Q
(wing equation, deri)-.275 E -.165(ve)-.275 G 2.75(df).165 G
(rom the one presented in [3] by setting R)-2.75 E .396 -.198
(TO = 4 * R)-.66 H(TT \(as it is)-.462 E(common practice\):)72 352.2 Q
/F3 11/Courier@0 SF 6.6(M=1)109.8 378.2 S
(/T = RTT_i * sqrt\(p\) * \(1 + 9*p * \(1 + 32*\(p\)^2\)\))-6.6 E F0
(where)72 420.8 Q 2.75(M=1)94 437.4 S(/T is proportional to the in)-2.75
E -.165(ve)-.44 G(rse of the throughput for the recei).165 E -.165(ve)
-.275 G 2.75(ru).165 G(nder consideration;)-2.75 E -.66(RT)94 454 S 2.75
(Ti).66 G 2.75(st)-2.75 G(he round trip time for the recei)-2.75 E -.165
(ve)-.275 G 2.75(ru).165 G(nder consideration;)-2.75 E 2.75(pi)94 470.6
S 2.75(st)-2.75 G(he loss rate for the recei)-2.75 E -.165(ve)-.275 G
2.75(ru).165 G(nder consideration, between 0 and 1.0;)-2.75 E
(and multiplying constants are omitted.)72 487.2 Q(The abo)72 513.2 Q
.33 -.165(ve e)-.165 H
(quation is accurate on a wide range of loss rates, and also co).165 E
-.165(ve)-.165 G(rs situations where).165 E(retransmission timeouts ha)
72 526.2 Q .33 -.165(ve a s)-.22 H
(igni\214cant impact on the throughput of the protocol.).165 E(Note tha\
t when p=0, the equation yields 1/T = M = 0. This does not constitute a\
 problem as we)72 552.2 Q(can still compare the M v)72 565.2 Q
(alues computed for dif)-.275 E(ferent recei)-.275 E -.165(ve)-.275 G
(rs to determine the ack).165 E(er)-.11 E 5.5(.A)-.605 G(lso note)-5.5 E
(that it is easier to compute M^2 instead of M, because the former does\
 not require the use of)72 578.2 Q(sqrt\(\).)72 591.2 Q(In future, dif)
72 617.2 Q
(ferent throughput equations may be substituted for this equation.)-.275
E(The requirement is)5.5 E(that the throughput equation be a reasonable\
 approximation of the sending rate of TCP for)72 630.2 Q
(conformant TCP congestion control.)72 643.2 Q(The parameters p and R)72
669.2 Q(TT need to be measured or calculated by a PGMCC implementation.)
-.66 E(The)5.5 E(measurement of R)72 682.2 Q(TT is speci\214ed in Secti\
on 2.5; the measurement of p is speci\214ed in Section 2.6.)-.66 E
(Rizzo/Iannaccone/V)72 769 Q(icisano/Handle)-.66 E 176.987(yS)-.165 G
(ection 2.4.)-176.987 E([P)5.5 E(age 9])-.165 E EP
%%Page: 10 10
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E/F1 11/Times-Bold@0 SF(2.5.)72
85 Q/F2 13/Times-Bold@0 SF -.52(RT)5.5 G 3.25(Tm).52 G(easur)-3.25 E
(ement)-.234 E F0(In PGMCC, the R)72 101.6 Q
(TT is measured by the sender making use of the timestamp \(or equi)-.66
E -.275(va)-.275 G(lent).275 E(information\) echoed back by each recei)
72 114.6 Q -.165(ve)-.275 G 2.75(ri).165 G 2.75(nf)-2.75 G
(eedback messages. Three procedures are possible)-2.75 E
(to measure the R)72 127.6 Q(TT)-.66 E 2.75(,a)-.814 G 2.75(sf)-2.75 G
(ollo)-2.75 E(ws. In no case is it required to ha)-.275 E .33 -.165
(ve c)-.22 H(lock synchronization between).165 E(sender and recei)72
140.6 Q -.165(ve)-.275 G(rs.).165 E F1 2.75(2.5.1. Explicit)72 179.6 R
-.198(Ti)2.75 G(mestamp).198 E F0(This \214rst technique relies on the \
transmission of a timestamp TS_j with each data pack)72 196.2 Q(et j.)
-.11 E(The recei)72 209.2 Q -.165(ve)-.275 G 2.75(rw).165 G
(ill record the most recently recei)-2.75 E -.165(ve)-.275 G 2.75(dt)
.165 G(imestamp, and will echo it back to the source)-2.75 E
(when generating an A)72 222.2 Q(CK or a N)-.44 E
(AK. If the feedback is delayed, the time elapsed between the)-.385 E(r\
eception of the timestamp and the generation of the feedback should be \
added to the echoed)72 235.2 Q(timestamp.)72 248.2 Q
(The sender computes the R)72 261.2 Q(TT by subtracting the recei)-.66 E
-.165(ve)-.275 G 2.75(dt).165 G(imestamp from the current v)-2.75 E
(alue of the)-.275 E(clock.)72 274.2 Q
(The resolution of the timestamp v)72 300.2 Q
(alue should be good enough for reasonable precision)-.275 E
(measurement of typical netw)72 313.2 Q(ork round trip times. If recei)
-.11 E -.165(ve)-.275 G(rs need to apply correction for delayed).165 E
(feedback, it is necessary that recei)72 326.2 Q -.165(ve)-.275 G
(rs kno).165 E 2.75(wt)-.275 G
(he resolution of the timestamp clock. A suggested)-2.75 E -.275(va)72
339.2 S(lue is 1ms.).275 E F1 2.75(2.5.2. Implicit)72 378.2 R(timestamp)
2.75 E F0 -.44(Wi)72 394.8 S(th this technique, the sender will record \
a timestamp TS_j for each transmitted data pack).44 E(et j,)-.11 E -.22
(bu)72 407.8 S 2.75(tt).22 G
(he timestamp will not be transmitted with the pack)-2.75 E(et itself.)
-.11 E(The recei)72 420.8 Q -.165(ve)-.275 G 2.75(rw).165 G
(ill record the most recently recei)-2.75 E -.165(ve)-.275 G 2.75(ds)
.165 G(equence number)-2.75 E 2.75(,a)-.44 G
(nd will echo it back to the)-2.75 E(source when generating an A)72
433.8 Q(CK or a N)-.44 E(AK.)-.385 E(The sender computes the R)72 446.8
Q(TT by looking up the timestamp associated with the sequence number)
-.66 E(recei)72 459.8 Q -.165(ve)-.275 G 2.75(di).165 G 2.75(nt)-2.75 G
(he feedback pack)-2.75 E
(et, and subtracting it from the current clock v)-.11 E(alue.)-.275 E
(If the feedback from the recei)72 485.8 Q -.165(ve)-.275 G 2.75(ri).165
G 2.75(sd)-2.75 G(elayed, as it is commonly the case for N)-2.75 E
(AKs, the recei)-.385 E -.165(ve)-.275 G(r).165 E(can compute, and send\
 back to the source, a correction term corresponding to the time elapse\
d)72 498.8 Q(between the reception of the timestamp and the generation \
of the feedback. The correction term)72 511.8 Q(will then be subtracted\
 by the sender in order to obtain the correct estimate of the R)72 524.8
Q(TT)-.66 E(.)-.814 E(This R)72 550.8 Q
(TT measurement technique is equi)-.66 E -.275(va)-.275 G
(lent to the pre).275 E(vious one, b)-.275 E(ut it sa)-.22 E -.165(ve)
-.22 G 2.75(ss).165 G(ome space in)-2.75 E(data pack)72 563.8 Q
(ets as the timestamp does not need to be sent e)-.11 E(xplicitly)-.165
E 2.75(.F)-.715 G(eedback pack)-2.75 E(ets might become)-.11 E(lar)72
576.8 Q(ger if the correction v)-.198 E(alue is transmitted e)-.275 E
(xplicitly; b)-.165 E(ut in man)-.22 E 2.75(yc)-.165 G
(ases, the sequence number)-2.75 E
(will already be present for other reasons \(e.g. A)72 589.8 Q(CK pack)
-.44 E(ets\), and where)-.11 E -.165(ve)-.275 G 2.75(rs).165 G
(pace is a concern the)-2.75 E
(sequence number and the correction term can be pack)72 602.8 Q
(ed in a single 32-bit w)-.11 E(ord without loss of)-.11 E(precision.)72
615.8 Q F1 2.75(2.5.3. Sequence)72 654.8 R(numbers)2.75 E F0
(This technique is the least precise, b)72 671.4 Q
(ut it does not rely on the presence of a high resolution clock)-.22 E
(on the nodes.)72 684.4 Q(The sender will not compute an)72 697.4 Q 2.75
(yt)-.165 G(imestamp, and just send data pack)-2.75 E
(ets with their sequence)-.11 E(number j.)72 710.4 Q(The recei)72 723.4
Q -.165(ve)-.275 G 2.75(rw).165 G(ill record the most recently recei)
-2.75 E -.165(ve)-.275 G 2.75(ds).165 G(equence number)-2.75 E 2.75(,a)
-.44 G(nd will echo it back to the)-2.75 E(Rizzo/Iannaccone/V)72 769 Q
(icisano/Handle)-.66 E 163.237(yS)-.165 G(ection 2.5.3.)-163.237 E([P)
5.5 E(age 10])-.165 E EP
%%Page: 11 11
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E(source when generating an A)72
85 Q(CK or a N)-.44 E(AK.)-.385 E(The sender computes the R)72 98 Q
(TT as the dif)-.66 E
(ference between the most recently sent sequence number)-.275 E
(and the sequence number recei)72 111 Q -.165(ve)-.275 G 2.75(df).165 G
(rom the A)-2.75 E(CK or N)-.44 E(AK pack)-.385 E(et.)-.11 E
(Note that in this case the R)72 137 Q(TT is not measured in seconds, b)
-.66 E(ut in "sequence numbers", which are)-.22 E(monotonically)72 150 Q
2.75(,b)-.715 G(ut not uniformly)-2.97 E 2.75(,i)-.715 G
(ncreasing with time. The tw)-2.75 E 2.75(om)-.11 G
(easurements are equi)-2.75 E -.275(va)-.275 G(lent if).275 E
(the sender transmits at a constant rate. When the data rate changes o)
72 163 Q -.165(ve)-.165 G 2.75(rt).165 G(ime \(as it is normally)-2.75 E
(happens, gi)72 176 Q -.165(ve)-.275 G 2.75(nt).165 G
(hat PGMCC controls the actual data rate\), then the "measured" R)-2.75
E(TT v)-.66 E(alues gro)-.275 E(w)-.275 E(with the actual transmit rate\
. This can in\215uence the correctness of the results when comparing)72
189 Q(tw)72 202 Q 2.75(om)-.11 G(easurement done o)-2.75 E -.165(ve)
-.165 G 2.75(rd).165 G(if)-2.75 E(ferent and only partially o)-.275 E
-.165(ve)-.165 G(rlapping time \(and sequence number\)).165 E(interv)72
215 Q(als where the transmit rate incurs a signi\214cant change.)-.275 E
/F1 11/Times-Bold@0 SF 2.75(2.5.4. Recommendations)72 254 R F0(Whene)72
270.6 Q -.165(ve)-.275 G 2.75(rp).165 G
(ossible, the measurement of the R)-2.75 E
(TT should be carried out using either e)-.66 E(xplicit or)-.165 E
(implicit timestamps, and by k)72 283.6 Q
(eeping track of the "correction term" \(the delay between data)-.11 E
(reception and feedback generation\).)72 296.6 Q(If the recei)72 322.6 Q
-.165(ve)-.275 G 2.75(rd).165 G(oes not ha)-2.75 E .33 -.165(ve a c)-.22
H(lock with a suitable resolution, the correction term might not be).165
E(present \(or be inaccurate\). In this case the timestamps recei)72
335.6 Q -.165(ve)-.275 G 2.75(db).165 G 2.75(yt)-2.75 G(he sender on N)
-2.75 E(AK pack)-.385 E(ets)-.11 E(might be in error)72 348.6 Q 2.75(,i)
-.44 G 2.75(nt)-2.75 G(he w)-2.75 E(orst case, by as much as the pack)
-.11 E(et interarri)-.11 E -.275(va)-.275 G 2.75(lt).275 G 2.75
(ime. This)-2.75 F(error will)2.75 E(normally not be present on A)72
361.6 Q(CK pack)-.44 E(ets, which are sent immediately)-.11 E 5.5(.A)
-.715 G(suitable correction)-2.75 E
(should be applied by the sender in order to a)72 374.6 Q -.22(vo)-.22 G
(id systematic errors.).22 E
(The measurement based on sequence numbers is less accurate, b)72 400.6
Q(ut also less sensiti)-.22 E .33 -.165(ve t)-.275 H 2.75(oe).165 G
(rrors due)-2.75 E(to the lack of the correction term. In f)72 413.6 Q
(act, the measurement error induced by the lack of the)-.11 E
(correction term can be at most one unit.)72 426.6 Q
(This suggests that, when the correction term is not)5.5 E -.22(av)72
439.6 S(ailable, measurements based on sequence numbers should be f)
-.055 E -.22(avo)-.11 G 2.75(ured. Simulations).22 F(ha)2.75 E -.165(ve)
-.22 G(sho)72 452.6 Q(wn that the ack)-.275 E
(er selection mechanism performs moderately better when the R)-.11 E(TT)
-.66 E(measurement is based on timestamps, b)72 465.6 Q
(ut performance is reasonably good also with)-.22 E
(measurements based on sequence numbers.)72 478.6 Q F1(2.6.)72 530.6 Q
/F2 13/Times-Bold@0 SF(Loss rate measur)5.5 E(ement)-.234 E F0
(The loss measurement in PGMCC is entirely performed by recei)72 560.2 Q
-.165(ve)-.275 G 2.75(rs. The).165 F(measurement results)2.75 E
(do not directly in\215uence the transmit rate, b)72 573.2 Q
(ut are only used for comparison purposes. As a)-.22 E
(consequence, the scheme is reasonably rob)72 586.2 Q(ust to dif)-.22 E
(ferent measurement techniques, as long as)-.275 E(the)72 599.2 Q 2.75
(ya)-.165 G(re not in\215uenced too strongly by single loss e)-2.75 E
-.165(ve)-.275 G(nts.).165 E
(The main method suggested for loss measurement is Exponentially Mo)72
625.2 Q(ving W)-.165 E(eighted A)-.88 E -.165(ve)-.814 G(rage).165 E
(\(EVMA\), which is formally equi)72 638.2 Q -.275(va)-.275 G
(lent to a single-pole digital lo).275 E 2.75(wp)-.275 G
(ass \214lter applied to a binary)-2.75 E
(signal x_i, where x_i = 1 if pack)72 651.2 Q
(et i is lost, x_i = 0 if pack)-.11 E(et i is successfully recei)-.11 E
-.165(ve)-.275 G(d.).165 E
(The loss rate p_i upon reception or detection of loss of pack)72 677.2
Q(et i is computed as)-.11 E/F3 11/Courier@0 SF
(p_i = c_p * p_{i-1} + \(1 - c_p \) * p_i)162.6 716.2 Q F0
(Rizzo/Iannaccone/V)72 769 Q(icisano/Handle)-.66 E 171.487(yS)-.165 G
(ection 2.6.)-171.487 E([P)5.5 E(age 11])-.165 E EP
%%Page: 12 12
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E(where the constant c_p betwee\
n 0 and 1 is related to the bandpass of the \214lter)72 85 Q 2.75(.E)
-.605 G(xperiments ha)-2.75 E -.165(ve)-.22 G(sho)72 98 Q(wn good perfo\
rmance with c = 500/65536, and computations performed with \214x)-.275 E
(ed point)-.165 E(arithmetic and 16 fractional digits.)72 111 Q
(As an alternati)72 137 Q .33 -.165(ve t)-.275 H 2.75(oE).165 G
(WMA, the technique used in TFRC [2] can be adopted. Simulations ha)
-2.75 E -.165(ve)-.22 G(sho)72 150 Q(wn a moderate impro)-.275 E -.165
(ve)-.165 G(ment in the ack).165 E
(er selection mechanism by measuring loss using the)-.11 E
(TFRC loss estimator)72 163 Q 2.75(,w)-.44 G(hich is ho)-2.75 E(we)-.275
E -.165(ve)-.275 G 2.75(rs).165 G(lightly more e)-2.75 E(xpensi)-.165 E
.33 -.165(ve t)-.275 H 2.75(oc).165 G(ompute than the EWMA loss)-2.75 E
(estimator in presence of pack)72 176 Q(et reordering.)-.11 E/F1 11
/Times-Bold@0 SF(2.7.)72 215 Q/F2 13/Times-Bold@0 SF -.234(Ti)5.5 G
(meouts).234 E F0(When a pack)72 244.6 Q
(et is transmitted, the sender schedules a timeout to pre)-.11 E -.165
(ve)-.275 G(nt stalls upon loss of A).165 E(CKs)-.44 E
(or disconnection of the A)72 257.6 Q 2.75(CKER. In)-.44 F(TCP)2.75 E
2.75(,w)-1.221 G(hich has a similar problem, the timeout v)-2.75 E
(alue is)-.275 E(computed by accumulating statistics \(SR)72 270.6 Q
(TT and R)-.66 E(TTV)-.66 E(AR\) on R)-1.485 E
(TT samples, starting from a)-.66 E(def)72 283.6 Q(ault initial v)-.11 E
(alue \(3s\) when no R)-.275 E(TT samples are a)-.66 E -.275(va)-.22 G
(ilable.).275 E(PGMCC can use a similar scheme to compute the timeouts,\
 remembering that upon A)72 309.6 Q(CKER)-.44 E
(changes \(which may be v)72 322.6 Q
(ery frequent\), the computation of SR)-.165 E(TT and R)-.66 E(TTV)-.66
E(AR must be restarted)-1.485 E(from the be)72 335.6 Q
(ginning, unless the sender decides to k)-.165 E
(eep state for at least a small number of recent)-.11 E(ack)72 348.6 Q
(ers.)-.11 E(Because the A)72 374.6 Q(CKER can lea)-.44 E .33 -.165
(ve t)-.22 H(he group without notifying the sender).165 E 2.75(,a)-.44 G
(fter a number of)-2.75 E(successi)72 387.6 Q .33 -.165(ve t)-.275 H
(imeouts the sender MUST force the election of a ne).165 E 2.75(wA)-.275
G 2.75(CKER. W)-3.19 F 2.75(er)-.88 G(ecommend this)-2.75 E(ne)72 400.6
Q 2.75(we)-.275 G(lection to be performed after tw)-2.75 E 2.75(os)-.11
G(uccessi)-2.75 E .33 -.165(ve t)-.275 H(imeouts.).165 E F1(2.8.)72
439.6 Q F2(Interaction with feedback suppr)5.5 E(ession schemes)-.234 E
F0(Se)72 469.2 Q -.165(ve)-.275 G(ral schemes are used by N).165 E
(AK-based multicast protocols to reduce the amount of feedback)-.385 E
(directed to)72 482.2 Q -.11(wa)-.275 G(rd the source and mak).11 E 2.75
(et)-.11 G(he protocol scale with lar)-2.75 E(ge populations of recei)
-.198 E -.165(ve)-.275 G 2.75(rs. Such).165 F
(schemes typically rely on randomly delaying N)72 495.2 Q
(AK generation, and suppressing pending N)-.385 E(AKs)-.385 E
(when an equi)72 508.2 Q -.275(va)-.275 G(lent N).275 E
(AK or a retransmission is heard; or)-.385 E 2.75(,i)-.44 G
(ntermediate nodes such as routers can)-2.75 E
(implement some form of feedback aggre)72 521.2 Q -.055(ga)-.165 G
(tion and \214ltering.).055 E(Such schemes might pre)72 547.2 Q -.165
(ve)-.275 G(nt N).165 E(AKs from potential A)-.385 E
(CKER candidates from reaching the source.)-.44 E(This \214ltering migh\
t impact the speed at which PGMCC selects the correct A)72 560.2 Q
(CKER, though initial)-.44 E -.165(ex)72 573.2 S
(perience from simulations seem to suggest that PGMCC beha).165 E
(viour is not se)-.22 E -.165(ve)-.275 G(rely af).165 E(fected by)-.275
E -.385(NA)72 586.2 S 2.75(Ks).385 G(uppression schemes.)-2.75 E F1
(2.9.)72 625.2 Q F2(Interaction with ECN)5.5 E F0
(PGMCC can use ECN noti\214cations in much the same w)72 654.8 Q
(ay as actual losses, and use such)-.11 E
(noti\214cations to control the throughput of the session.)72 667.8 Q
(At the recei)72 693.8 Q -.165(ve)-.275 G .88 -.44(r, E).165 H(CN-mark)
.44 E(ed data pack)-.11 E(ets can be considered as lost pack)-.11 E
(ets for the purpose of)-.11 E(loss rate computation and A)72 706.8 Q
(CK/N)-.44 E(AK generation.)-.385 E(If the A)5.5 E(CKER sends an A)-.44
E(CK for ECN-mark)-.44 E(ed)-.11 E(pack)72 719.8 Q(ets, that A)-.11 E
(CK MUST reports that the pack)-.44 E(et being ack)-.11 E(ed that w)-.11
E(as ECN mark)-.11 E 2.75(ed. Similarly)-.11 F(Rizzo/Iannaccone/V)72 769
Q(icisano/Handle)-.66 E 171.487(yS)-.165 G(ection 2.9.)-171.487 E([P)5.5
E(age 12])-.165 E EP
%%Page: 13 13
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E(the A)72 85 Q
(CKER must indicate in the A)-.44 E(CK pack)-.44 E(et')-.11 E 2.75(sr)
-.605 G(ecei)-2.75 E -.165(ve)-.275 G 2.75(dp).165 G(ack)-2.75 E
(ets bitmap that the pack)-.11 E(et w)-.11 E(as ECN-)-.11 E(mark)72 98 Q
(ed, or that the pack)-.11 E(et w)-.11 E(as lost.)-.11 E 1.76 -.88(We n)
72 124 T(ote that to support use of the ECN nonce, the A).88 E(CK pack)
-.44 E(et')-.11 E 2.75(sr)-.605 G(ecei)-2.75 E -.165(ve)-.275 G 2.75(dp)
.165 G(ack)-2.75 E(ets bitmap w)-.11 E(ould)-.11 E(require tw)72 137 Q
2.75(ob)-.11 G(its per pack)-2.75 E(et being reported.)-.11 E/F1 11
/Times-Bold@0 SF(3.)72 176 Q/F2 14/Times-Bold@0 SF(Pr)5.5 E(ocedur)-.252
E(es - Sender)-.252 E F0(The follo)72 192.6 Q
(wing pseudo-code speci\214es the complete beha)-.275 E
(viour of the sender in PGMCC.)-.22 E/F3 8/Courier@0 SF(initialization:)
72 231.6 Q 4.8(T=1;W=1; /)81.6 244.6 S 4.8(*i)-4.8 G
(nitialize window and number of tokens */)-4.8 E(RETRY = 0 ;)81.6 257.6
Q(/* number of consecutive timeouts so far)33.6 E(*/)9.6 E 4.8(<i)81.6
270.6 S(nitialize p, RTT for acker to default values >)-4.8 E
(ACKER = NO_ACKER; /* no acker is known)81.6 283.6 Q(*/)105.6 E 4.8(<i)
81.6 296.6 S(nitialize sequence numbers >)-4.8 E(QUEUED = 0;)81.6 309.6
Q(/* packets waiting to be transmitted)33.6 E(*/)28.8 E
(on transmission request:)72 335.6 Q(send_packet\(\) ;)81.6 348.6 Q
(on timeout expiration :)72 374.6 Q 4.8(T=1;W=1; /)81.6 387.6 S 4.8(*i)
-4.8 G(nitialize window and number of tokens */)-4.8 E
(if \(RETRY < RETRY_MAX\))81.6 400.6 Q(RETRY = RETRY + 1)91.2 413.6 Q
(else)81.6 426.6 Q(ACKER = NO_ACKER ; /* old acker is not valid anymore)
91.2 439.6 Q(*/)28.8 E(send_packet\(\) ;)81.6 452.6 Q
(on ACK/NAK reception from receiver I :)72 478.6 Q 4.8(<c)81.6 491.6 S
(ompute p and RTT for source of this ACK, see Sec. 2.5 and 2.6 >)-4.8 E
(RETRY = 0 ;)81.6 504.6 Q(if \(ACKER == NO_ACKER\) { /* select current \
as acker is no other known */)81.6 517.6 Q(ACKER = I ;)91.2 530.6 Q 4.8
(T=T+1;)91.2 543.6 S(})81.6 556.6 Q(if \(ACKER != I\))81.6 569.6 Q 4.8
(<s)91.2 582.6 S(elect acker according to Sec. 2.3 > ;)-4.8 E(else {)
81.6 595.6 Q(<update acker statistics \(p_ACKER, RTT_ACKER\) >)91.2
608.6 Q(if \(packet_type == ACK\) {)91.2 621.6 Q 4.8(<u)100.8 634.6 S
(pdate_window according to Sec.2.2 >)-4.8 E(send_packet ;)100.8 647.6 Q
(if \(ack_pending\))100.8 660.6 Q(update_timeout ;)110.4 673.6 Q(})81.6
686.6 Q(})72 699.6 Q F0(Rizzo/Iannaccone/V)72 769 Q(icisano/Handle)-.66
E 179.737(yS)-.165 G(ection 3.)-179.737 E([P)5.5 E(age 13])-.165 E EP
%%Page: 14 14
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E/F1 8/Courier@0 SF
(send_packet\(\) {)72 85 Q(if \(QUEUED > 0 && T >= 1\) {)81.6 98 Q 4.8
(<t)91.2 111 S(ransmit one packet >)-4.8 E 4.8(T=T-1;)91.2 124 S
(QUEUED = QUEUED - 1 ;)91.2 137 Q(})81.6 150 Q
(if \( <there are unacked packet> \))81.6 163 Q
(<set a timeout, see Sec.2.7 >)91.2 176 Q(})72 189 Q/F2 11/Times-Bold@0
SF(4.)72 254 Q/F3 14/Times-Bold@0 SF(Pr)5.5 E(ocedur)-.252 E
(es -- Recei)-.252 E -.14(ve)-.14 G(r).14 E F0(The follo)72 270.6 Q
(wing pseudo-code speci\214es the complete beha)-.275 E
(viour of the recei)-.22 E -.165(ve)-.275 G 2.75(ri).165 G 2.75(nP)-2.75
G(GMCC.)-2.75 E 2.75(Ar)72 296.6 S(ecei)-2.75 E -.165(ve)-.275 G 2.75
(ro).165 G(nly transmits an A)-2.75 E(CK pack)-.44 E(et when it recei)
-.11 E -.165(ve)-.275 G 2.75(sad).165 G(ata pack)-2.75 E
(et for which the recei)-.11 E -.165(ve)-.275 G 2.75(ri).165 G(s)-2.75 E
(designated as the A)72 309.6 Q(CKER by the data pack)-.44 E(et itself.)
-.11 E 2.75(Ar)5.5 G(ecei)-2.75 E -.165(ve)-.275 G 2.75(rc).165 G
(an transmit a N)-2.75 E(AK pack)-.385 E(et after it)-.11 E
(has detected that a data pack)72 322.6 Q
(et is missing and a suitable delay has passed, as dictated by the)-.11
E(feedback suppression rules of the protocol in use.)72 335.6 Q
(The data pack)72 361.6 Q(et contains ackno)-.11 E
(wledgement status about the most recent 32 sequence numbers)-.275 E
(kno)72 374.6 Q(wn to the recei)-.275 E -.165(ve)-.275 G -.605(r.).165 G
F1(on initialization/session setup:)72 413.6 Q 4.8(<i)81.6 426.6 S
(nitialize state variables and ACK bitmap >)-4.8 E
(on DATA packet reception:)72 452.6 Q 4.8(<u)81.6 465.6 S
(pdate p measurement according to Sec.2.6 >)-4.8 E 4.8(<r)81.6 478.6 S
(ecord timestamp and packet reception time >)-4.8 E
(if \(ACKER == this_node\) {)81.6 491.6 Q 4.8(<s)91.2 504.6 S
(end an immediate ACK >)-4.8 E(})81.6 517.6 Q
(if \( <some data packet is missing and unacknowledged > \))81.6 530.6 Q
4.8(<s)91.2 543.6 S(chedule a timeout for NAK transmission >)-4.8 E
(on NAK reception:)72 569.6 Q 4.8(<s)81.6 582.6 S
(uppress any pending NAK transmission for the sequence)-4.8 E
(number indicated in the NAK >)91.2 595.6 Q(on timeout:)72 621.6 Q
(if \( < there are missing and unacknowledged packets > \) {)81.6 634.6
Q 4.8(<s)91.2 647.6 S
(end a NAK for one or more of the missing packets >)-4.8 E 4.8(<m)91.2
660.6 S(ark such packets as acknowledged >)-4.8 E
(if \( <some data packet is missing and unacknowledged > \))91.2 673.6 Q
4.8(<s)100.8 686.6 S(chedule a timeout for NAK transmission >)-4.8 E(})
81.6 699.6 Q F0(Rizzo/Iannaccone/V)72 769 Q(icisano/Handle)-.66 E
179.737(yS)-.165 G(ection 4.)-179.737 E([P)5.5 E(age 14])-.165 E EP
%%Page: 15 15
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E/F1 11/Times-Bold@0 SF(5.)72 85
Q/F2 14/Times-Bold@0 SF(Security Considerations)5.5 E F0
(PGMCC is not a transport protocol in its o)72 101.6 Q(wn right, b)-.275
E(ut a congestion control mechanism that is)-.22 E
(intended to be used in conjunction with a transport protocol.)72 114.6
Q(Therefore security primarily needs)5.5 E
(to be considered in the conte)72 127.6 Q(xt of a speci\214c transport \
protocol and its authentication mechanisms.)-.165 E
(Congestion control mechanisms can potentially be e)72 153.6 Q
(xploited to create denial of service.)-.165 E(This may)5.5 E
(occur through spoofed feedback.)72 166.6 Q(Thus an)5.5 E 2.75(yt)-.165
G(ransport protocol that uses PGMCC should tak)-2.75 E 2.75(ec)-.11 G
(are)-2.75 E(to ensure that feedback is only accepted from the recei)72
179.6 Q -.165(ve)-.275 G 2.75(ro).165 G 2.75(ft)-2.75 G(he data.)-2.75 E
(The precise mechanism to)5.5 E(achie)72 192.6 Q .33 -.165(ve t)-.275 H
(his will ho).165 E(we)-.275 E -.165(ve)-.275 G 2.75(rd).165 G
(epend on the transport protocol itself.)-2.75 E(In addition, congectio\
n control mechanisms may potentially be manipulated by a greedy recei)72
218.6 Q -.165(ve)-.275 G(r).165 E(that wishes to recei)72 231.6 Q .33
-.165(ve m)-.275 H(ore than its f).165 E(air share of netw)-.11 E
(ork bandwidth.)-.11 E 2.75(Ar)5.5 G(ecei)-2.75 E -.165(ve)-.275 G 2.75
(rm).165 G(ight do this by)-2.75 E
(\214rst reporting in\215ated loss and R)72 244.6 Q
(TT samples, in order to get selected as the A)-.66 E(CKER, and then)
-.44 E(generating A)72 257.6 Q
(CK at the desired rate \(including possibly claiming to ha)-.44 E .33
-.165(ve r)-.22 H(ecei).165 E -.165(ve)-.275 G 2.75(dp).165 G(ack)-2.75
E(ets that in)-.11 E -.11(fa)72 270.6 S
(ct were lost due to congestion\).).11 E(Possible defenses ag)5.5 E
(ainst such a recei)-.055 E -.165(ve)-.275 G 2.75(rc).165 G
(ould be based on the)-2.75 E(sender v)72 283.6 Q
(erifying the correctness of the loss and R)-.165 E
(TT samples supplied by the recei)-.66 E -.165(ve)-.275 G 2.75(r\().165
G(by)-2.75 E(comparing them with information deri)72 296.6 Q -.165(ve)
-.275 G 2.75(df).165 G(rom the incoming stream of A)-2.75 E
(CKs\). Also, the sender)-.44 E
(could include some form of nonce that the recei)72 309.6 Q -.165(ve)
-.275 G 2.75(rm).165 G(ust feed back to the sender to pro)-2.75 E .33
-.165(ve r)-.165 H(eceipt.).165 E(Ho)72 322.6 Q(we)-.275 E -.165(ve)
-.275 G .88 -.44(r, t).165 H(he details of such a nonce w).44 E
(ould depend on the transport protocol, and in particular on)-.11 E
(whether the transport protocol is reliable or unreliable.)72 335.6 Q F1
(6.)72 374.6 Q F2 -.7(Au)5.5 G(thors' Addr).7 E(esses)-.252 E F0
(Luigi Rizzo)80.25 391.2 Q(luigi@iet.unipi.it)80.25 404.2 Q
(Dip. Ing. dell'Informazione,)80.25 417.2 Q(Uni)80.25 430.2 Q 1.43 -.715
(v. d)-.275 H 2.75(iP).715 G(isa)-2.75 E
(via Diotisalvi 2, 56122 Pisa, Italy)80.25 443.2 Q(Gianluca Iannaccone)
80.25 469.2 Q(gianluca@sprintlabs.com)80.25 482.2 Q(Sprint A)80.25 495.2
Q(TL)-1.221 E 2.75(1A)80.25 508.2 S(drian Ct., Burling)-2.75 E
(ame, CA, USA)-.055 E(Lorenzo V)80.25 534.2 Q(icisano)-.66 E
(lorenzo@cisco.com)80.25 547.2 Q(cisco Systems, Inc.)80.25 560.2 Q
(170 W)80.25 573.2 Q(est T)-.88 E(asman Dr)-.88 E(.,)-.605 E
(San Jose, CA, USA, 95134)80.25 586.2 Q(Mark Handle)80.25 612.2 Q(y)
-.165 E(mjh@aciri.or)80.25 625.2 Q(g)-.198 E(ICSI,)80.25 638.2 Q
(1947 Center St, Berk)80.25 651.2 Q(ele)-.11 E 1.43 -.715(y, C)-.165 H
(A, USA, 94704).715 E F1(7.)72 690.2 Q F2(Ackno)5.5 E(wledgments)-.14 E
F0 1.76 -.88(We w)72 706.8 T(ould lik).77 E 2.75(et)-.11 G 2.75(oa)-2.75
G(ckno)-2.75 E
(wledge feedback and discussions on equation-based congestion control)
-.275 E(with a wide range of people, including members of the Reliable \
Multicast Research Group, the)72 719.8 Q(Rizzo/Iannaccone/V)72 769 Q
(icisano/Handle)-.66 E 179.737(yS)-.165 G(ection 7.)-179.737 E([P)5.5 E
(age 15])-.165 E EP
%%Page: 16 16
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E(Reliable Multicast T)72 85 Q
(ransport W)-.385 E(orking Group, and the End-to-End Research Group.)
-.88 E([1] Bradner)72 140.6 Q 2.75(,S)-.44 G(., K)-2.75 E .33 -.165
(ey w)-.275 H(ords for use in RFCs to Indicate Requirement Le).055 E
-.165(ve)-.275 G(ls \(IETF RFC 2119\)).165 E(http://www)72 153.6 Q
(.rfc-editor)-.715 E(.or)-.605 E(g/rfc/rfc2119.txt)-.198 E([2] Flo)72
179.6 Q(yd, S., Handle)-.11 E 1.43 -.715(y, M)-.165 H(., P).715 E(adh)
-.165 E(ye, J., W)-.055 E(idmer)-.44 E 2.75(,J)-.44 G
(., "Equation-Based Congestion Control for)-2.75 E
(Unicast Applications", A)72 192.6 Q
(CM SIGCOMM 2000, Stockholm, Aug. 2000)-.44 E([3] P)72 218.6 Q(adh)-.165
E(ye, J. and)-.055 E(Firoiu, V)5.5 E 2.75(.a)-1.419 G(nd T)-2.75 E -.275
(ow)-.88 G(sle).275 E 1.43 -.715(y, D)-.165 H 2.75(.a).715 G(nd K)-2.75
E(urose, J., "Modeling TCP Throughput: A)-.165 E
(Simple Model and its Empirical V)72 231.6 Q(alidation", Proc A)-1.221 E
(CM SIGCOMM 1998.)-.44 E([4] Mankin, A., Romano)72 257.6 Q 1.43 -.715
(w, A)-.275 H(., Brander).715 E 2.75(,S)-.44 G(., P)-2.75 E(axson, V)
-.165 E(., "IETF Criteria for Ev)-1.419 E(aluating Reliable)-.275 E
(Multicast T)72 270.6 Q
(ransport and Application Protocols," RFC2357, June 1998.)-.385 E([5] R\
izzo, L., "pgmcc: a TCP-friendly single-rate multicast congestion contr\
ol scheme", A)72 296.6 Q(CM)-.44 E(SIGCOMM 2000, Stockholm, Aug.2000)72
309.6 Q([6] Schulzrinne, H., Casner)72 335.6 Q 2.75(,S)-.44 G
(., Frederick, R., Jacobson, V)-2.75 E(., "R)-1.419 E(TP: A T)-.66 E
(ransport Protocol for Real-)-.385 E -.385(Ti)72 348.6 S
(me Applications", RFC 1889, Jan 1996.).385 E([7] Speakman, T)72 374.6 Q
(., Cro)-.814 E(wcroft, J., Gemmell, J., F)-.275 E
(arinacci, D. , Lin, S., Leshchiner)-.165 E 2.75(,D)-.44 G(., Luby)-2.75
E 2.75(,M)-.715 G(.,)-2.75 E(Montgomery)72 387.6 Q 2.75(,T)-.715 G 2.75
(.,R)-3.564 G(izzo, L., T)-2.75 E(weedly)-.88 E 2.75(,A)-.715 G
(., Bhaskar)-2.75 E 2.75(,N)-.44 G(., Edmonstone, R., Sumanasek)-2.75 E
(era, R.,)-.11 E -.66(Vi)72 400.6 S(cisano, L., PGM Reliable T).66 E
(ransport Protocol Speci\214cation, RFC 3208, December 2001.)-.385 E
(rfc3208.txt also a)72 413.6 Q -.275(va)-.22 G
(ilable at ftp://ftp.rfc-editor).275 E(.or)-.605 E
(g/in-notes/rfc3208.txt)-.198 E/F1 11/Times-Bold@0 SF(8.)72 478.6 Q/F2
14/Times-Bold@0 SF(Full Copyright Statement)5.5 E F0(Cop)72 495.2 Q
(yright \(C\) The Internet Society \(2000\).)-.11 E(All Rights Reserv)
5.5 E(ed.)-.165 E(This document and translations of it may be copied an\
d furnished to others, and deri)72 511.8 Q -.275(va)-.275 G(ti).275 E
.33 -.165(ve w)-.275 H(orks).055 E(that comment on or otherwise e)72
524.8 Q
(xplain it or assist in its implementation may be prepared, copied,)
-.165 E(published and distrib)72 537.8 Q
(uted, in whole or in part, without restriction of an)-.22 E 2.75(yk)
-.165 G(ind, pro)-2.75 E(vided that the)-.165 E(abo)72 550.8 Q .33 -.165
(ve c)-.165 H(op).165 E(yright notice and this paragraph are included o\
n all such copies and deri)-.11 E -.275(va)-.275 G(ti).275 E .33 -.165
(ve w)-.275 H(orks.).055 E(Ho)72 563.8 Q(we)-.275 E -.165(ve)-.275 G .88
-.44(r, t).165 H(his document itself may not be modi\214ed in an).44 E
2.75(yw)-.165 G(ay)-2.86 E 2.75(,s)-.715 G(uch as by remo)-2.75 E
(ving the)-.165 E(cop)72 576.8 Q(yright notice or references to the Int\
ernet Society or other Internet or)-.11 E -.055(ga)-.198 G(nizations, e)
.055 E(xcept as)-.165 E(needed for the purpose of de)72 589.8 Q -.165
(ve)-.275 G(loping Internet standards in which case the procedures for)
.165 E(cop)72 602.8 Q
(yrights de\214ned in the Internet languages other than English.)-.11 E
(The limited permissions granted abo)72 619.4 Q .33 -.165(ve a)-.165 H
(re perpetual and will not be re).165 E -.22(vo)-.275 G -.11(ke).22 G
2.75(db).11 G 2.75(yt)-2.75 G(he Internet)-2.75 E
(Society or its successors or assigns.)72 632.4 Q
(This document and the information contained herein is pro)72 649 Q
(vided on an "AS IS" basis and THE)-.165 E
(INTERNET SOCIETY AND THE INTERNET ENGINEERING T)72 662 Q
(ASK FORCE DISCLAIMS)-1.023 E(ALL W)72 675 Q
(ARRANTIES, EXPRESS OR IMPLIED, INCLUDING B)-1.32 E(UT NO)-.11 E 2.75
(TL)-.44 G(IMITED T)-2.75 E 2.75(OA)-.198 G(NY)-2.75 E -1.32(WA)72 688 S
(RRANTY THA)1.32 E 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)-2.75 E
(ANY RIGHTS OR ANY IMPLIED W)72 701 Q(ARRANTIES OF MERCHANT)-1.32 E
(ABILITY OR FITNESS)-1.023 E(FOR A P)72 714 Q(AR)-1.012 E
(TICULAR PURPOSE.")-.66 E(Rizzo/Iannaccone/V)72 769 Q(icisano/Handle)
-.66 E 179.737(yS)-.165 G(ection 8.)-179.737 E([P)5.5 E(age 16])-.165 E
EP
%%Page: 17 17
%%BeginPageSetup
BP
%%EndPageSetup
/F0 11/Times-Roman@0 SF(INTERNET)72 49 Q 66.092(-DRAFT Expires:)-1.012 F
(December 2002)2.75 E(June 2002)116.406 E(Rizzo/Iannaccone/V)72 769 Q
(icisano/Handle)-.66 E 179.737(yS)-.165 G(ection 8.)-179.737 E([P)5.5 E
(age 17])-.165 E EP
%%Trailer
end
%%EOF