ࡱ> ts@=a$Yo( ('0%Ӓ!q/xmlSUvʀG 0 Ą\`LdCf%eX?0)EcQ cD" LX"!A1ѐLL9]{mGqw={܏@/aa  oAlU `r {qTh0@H(nw#BBװ)$)v]B [ <\CV| $./Ny,o Q>.iGuXt|Gp?]Q;Svr4̵*{)U D8 a\$]p:1z @9ŀ 3A C? /4Ʊ^c:kh 87b`ϘՀj`xԬ0lLS-G;zAdGP7=boz&{)|y@zu3ҕLCKZDklɷn`S\LG.}:xu$BzuE̕G=?lW=5G_7{1hTҷF,څ鄓>6GG-oAӭL(cu+y|ͷVItX l)|lEIT0W8n31bzy(%]`S!Nb`sV g9VV W'پ|:m_J Ph!BB $_oJW'm%~Ѣ6g lIt"%gLۨJMS'%d7g1=縵KdXôaY}KE .bnͶ.?.gu5F ?p]W, Ї+/227LU/o+r/Dx zH hTYJ;=^8:2*2"27p+$a3)R9,p,ЦoAv;aT #^M4Gs>]8hj6R~ǂ|~&TL.]9q(MO=#KI;} bTw 'p{mh / h}"NyEȳԃF3Xsco^) H*6N+ۓݭȍ|˹;XPbsLRN+R9lE9nގb{uڎQ1zfHxU̧ح|t՝Rn^.w\N/g_z_{/C6$h5.e#=0{W@]? >VJNd|\]*߭.~+'둩ĞM+į|(fr}2 , TLf9Q:X57ֹğ,326?xN6(Gx$(){Z C\t'o#7H=#H I>60 A i XF|HX/A p, [߆!lCp9@%l'@]p 6 ȆM'^c 9'@@+'vQ?.Efl0d|\/`H ue'& MR LPW(s FBd0}PP+^$nHLn$S U%H )]TLg@D`º`~~gC1wQ  L/XG}D 'y4Ў*C.Ib>34Po48Sz_!O )LsXVGR/Te6_xc|f{|-zy%pf/vOƸ ͏&zyR+:*&L ՛hS,8>@Poc~D_WP&nHM|.u&#r?aRT@_kw$m~N/X&`O<NeC,LĈ `9K+g b̖ G6 R@KC0@6YkKHŶ5Y!0@ ol2O%6-][`^?щu#v>7N9SLEÏ"-e?%A=)#ݴ?D7K'orq,kKY&L? NPbFpѿ|7kKY5ќL 7[{%>-%"@7@5#L/O"Lz@`Tl{ cz? @ 9GdBB) ƄՓ%p 8@Rdn +.z(i 0,A@7_ R&kH@iFC6{|a\kg.S IFH(@JMd14לw*I.} S( A qJR$w#*=La'@VNU ܍3ճ8@:_O Eً@Ag' L哥4?MX6UJ-c$ 1e.{V)5,1HVꅐI*$Zhj?@:Y(CimC (ȾK1OJ-XFZ\ >)R֕ӂ Sx&@:,,~ '˺--9?8*B& Z})>)DDn⯅͂p݆S`5G(·6Ɓk , Wt"<#W؃؆ݴ$,6StD :U  9Dz+I دIJB(U'],g{_ Woj68{%W, B A2)Xݰd6=ts3Ó?}V_ҁ:@P\@hi_C5.9n<#_А`a UN]Jݐ`eo/(Bţ \n pjp3\z@[˨z>@c#&ik'x3דHF`2X_@ ٫s29@)W@p MH1LFMuK,Jlũ0&\o&{.K삊 F]O #XddLK_`$L$,[慰O,rEb `}RA?!hi?l<]ict3_*jB\VTPX-7FӋ#VJ0HMTNwGǜ6B#AgHN #JX1#Mv["RWcv_bWav(~s~|N~ί,;Qp?~"{I\exG?6NIT{*~ 1*wڡsZ%ukr{?|>ڜlׄqyuIg>І>p{RA_m6#N`P6oaH`]!|{4s>« B %@#n=Tk` $ / &2P *Ra o2u[\i왋.h&Aۮ?6n~zF:-(;[ReEzўeI# 0%UF. X]RVFwhVGY@"8 peae-d`  |B`N @K.@>(13?ql@(: a'| 0rlpBXgFzVtl*7 ',KC;[pؘm8)X`y܂_2̯ 4!d~VhC6`N(+h3%SPaAk8l^ӀPk< " »U d@FB,U !F AEH\FȖ# mt!ZCo0B4ʹM*Y_"q}?Q92es dI!ckС|oֆmY5f~kS9zP#\8 q#GoBl< "Kۊ<53AB2e$n" *d"$,8AAB&lF$Zw4BGΤBύ2Hd/B"j׈T1N3"]Ŭ+DNhڐeZ+#!_BT!q B6M!AAABB#b' !Ki b% !DABV$$ Bfql$6"ABVf$2"BȀ df#؈ !Y ]b#Q yoDs)Lr{dJeAI0F6 ـ!1Ty~+$+g AB)dT@{]5!!f4#g`YNO5؅\nס"k? WABhF dӚ087#B VC0R|Z-/b! B)?7u1 tyƻFB{AUq~Bq 6oM.NtPP\#Y'!%$2Q٣ͩKdW'Z/Z8]}6 }2$.!%;BR =dj! qvϕw/_F+DZ+S&{zx鑫>\ &x^Jn6"Kws:}kdԩ.5Eȣ8=vl29Ձ|v"*BLUg,Jiꮁ|yNE/"c:#H'b6$s)!ۑ{I})M"tn1 99 !1OGTP`*D,@>nF9e WlA,-ˬDl dqy6!a@Dƍ>z~P&9NVB]$< 1v9}/ĉB#cH:> u8!G-! &_"#&L4ky+rA:FǯY|8I'.)I~~]#c%nhm8]șW Fe=f@ȥve2?} qt! 3 B$x BZ7*(Ys2U[^<!{5*KT!d݀FH!Sp}B>"Sp}B)B8 *B{qNAȀt2;!b*BlwOBR@ d2b*Bq B" "boa8T?B0_NYCo'W /eLdh,>elB[DLr_^٪wmZb{Q?AH__w,x\[d1;eJ@HD&`}B/GBBP!"V2f IBFR3ABH[lDχl.PmA-!!!!!!!) i-~dz&bABK.gѪ^g!M 6! ڝ !a<!" !!!!!!!!ABBBBBBpB@{!!ABBBzhCUAH!SC"xVC''_A!!!!! t7bb+×ƣw+5w#H#}U몐ƪ|/E?mۄqP!J؆*l!akP=*2F|BϡE!DcBF񽢞8!J^&g%=3*٫W^ij] ēl+gufv'W Bҟ+{ !ɪ'R8uy򧛳 d<5<#}>7Ze7Bvo^OGF2: xt)gG-!d9)uȋ[gC\xTEAAAAAAAA!!!!!!!!?XIENDB`@=f2䘳lI$|w7 %exmlUpvo{^rHփ@O-X4Ј?Pmkʕӈ%?H$F ?BLDi& chhIy7ݻ&s;of޼y3۝=b% ո` ܺq18|LL(x= dPRҐ#46݊βX] x/x$j hXQ^8c|B%&E(EX4!;nG\%e1Ա"Ec^A ,f9s1GUS1^cN26֭a.A؄x1/lN)b] -XWcB0c"XA|Wv$ql5ͯQ3S(+pyVGCsi ٙy vC.$@UP񾥻UyGg]W< Mݝ {cl:u(:<1\Zn@ɾ waƊ/}O~2._߶W&|>0þ\쓾'{K}?7|~;F2'>ۃ!|o!10^4否mAw$I6Ch9m$Od֟͂~'3˓K?aƋ! /f|Ot% >G*Gq#^͙f8i"o*g9U9Y;OKo/O+z(??4X1ԑf3҈E.we ʏsRԤ,3\ U}.N+m꫷`zo>q [d~tEf'pmm9k1gx=F{r-t\ %! xZ)~ǹȟBiӢ-kР66&k6lO~`>Urr,ZzRlK 堀m x9Vdg½h7f~^SkM5L y"kWRd [ v5,:-d8I߼ 3/ B1~h#[;΄`[KկC~By(2ok|#$\ F!7oe8}-#.K5>}6IcA=v5IBpJܙLzT|cEK[K!ZJ\aU/=vѳi,(]Vzl{k=ڙ-r9S;{ ,%xZ0D>2e _F:f&/gְzhu7޸X{em]#kSWkK5/z!\y?p7\r6Ӡ#t#Rsȣ>/SSϙ+ԖbӖB#T1纎k9?%Q\ܔxwc'0n" ]xT5WxGPNG  IHDR ͥ^sRGB pHYsjj|l~W6g@@@@eQFeQ&P&P&P&P +BNџ]9+'3(2G>b)L5 /ҭHF7M-eּg4ۖ;T(Kbt4ٶƭTkceIY~xhzɵ݁Zyn ʋSYԱ-%c(* I;@y- ?G\50ڧ|Y_.e_H9SDO? EmvG֊H_ݻáZ@_=\eUWҒr*WNw燈֪RN݅E:wƉsDa]LԧzvW, %a3)R!R8DlU}UJ/C8S )NNgފ(YlarVb+ʱ2դg_&Y۳`vyf3s3P(9O}N'-fm{;S_^Tb&M}yY@/0kw SjwuyvR=QvÛ)=7fVnjQb1X-2eeí1[(&IP^!~*]' kOM U=kCy&sXUfp= UY"׺(/:.z^nݛpA)+Lӧkef){*5WyfϽ^Pv%̓Bys0"(;+oό2C 'ffz͌a7fFy/f,bFyGi9^\y)g l0Rr?*r{Z SFV].i)@ًNCj53gx1Q*;$'^3يϠY'oko(!{2v$q/^^GR{fL}y fweL|Pb2==NyˊP \&e+f8g~6\95.QyRfS'ҵ@9D̼H:'+;gK1;g+efl K1_gfUgOfbh( [2.4iܤ}9wHJM`F^bN5r3~oƄܚw 9RVחbF 1[EDʡ(+SpV1M`pwZQN$*̉`9<TY/(W.OWFA?rվ9"P>T222222(2(2(222222(2ʄromܟe |EI1/WYm19ڞ!Pm _$6Doy-X6ۉ7|=#؋勘AꚆX[ *=çi-Ln|[CY,fB,\\EYN(᷊R]AY@W8ί>8#^~Qr*k"sl9v;.\8Ts|Am22222222EhSΣ|wm(9rG<'w!E\\yCUP>myOu>R*ʇRΟ~J#K]b%|xy"[ kEMu}i}P>Fn7vygka8~cߏ,B e Q&P&P&P&ᾅ?tlIENDB`@=  $RW.œ)|w7X $r xZh?o^~Yk|mpmck{Z-’6 KEi/mH}j*9F.J0AjKc΍c0Av>r.{={3 Fހ 5=N4@TOlSM7+p?:kV;Ueƀ4#Hl;-)Zl\y*/o.-M5ZXwMlyjjKm9_F i۩8"s6:em&閼.P>v}%umBU%'(ӟV}g~a2Ou%^WS7uʺdPI;_}Nb2s6nY tq;9?_2Ç-Gގo^Tga~|xл~PzW[!s22~-څBQv5OM(#N>+0xxm"qzg3woW@ ԰7G{ |'ZzlŗӖ:v\'ܸغ[7. #/rTq8@5Pʷ=/#}rNwN*sNC1締-P:îyVmg.xӼ4RysmSO߲mc, e#rqv~?fedyʷ:n_DZ/c݌^͓ zRͭᩲl{O&l.EkK"k. 3hW=Jsf;kPC`dgNp=/?~>&v.8ءbuf?1(ˬ"02{\,pzѭb`M?쯰?~޺+Q`b1\Xݬ-0'\uք`(J7عXtWv&/0|U0ˀ~QW;+og͠?pC,FA1&η6h} )nك>>ޥc0Q,o4=48 ?ܫ`0F;]G#9*0'+{XUαS~=8㋊Y# YAay'Keg*430>}Ϯ:(^ןOzoxox^Ws^}Wկxu1+x͑ 2:՟2nu>V3YUWjy3˳@j8׌<'.O.HomUQ|ݞOOWWQ'cY]esdnq?M}# q Qq 9_ك|_-f0(xx_1(nYsD`dgN*n˂^# qbαg=8}8a1~}[qj7Sӟ|cM?0>(nQ# Oo )yo ŭ5'FAvXWGX^Hcyv&or1>HЎy1>%ߎ鞻ĭSӵ֕Uny~Ss݋9Z ]`7=pKP/'ko%G}xu&CW^Rα_ ^Hϱ]fHyjg<úi.tpz'[7W6eiYmRc^Rv\VZݺR5JQzθ^;f*>.7MIxJŤZ9.V[QkMdu^Ϸy~R1VWؾG±Ex3鵶ӶLWc}G]6(   8mailto:anup.ghosh@in.ibm.com8mailto:anup.ghosh@in.ibm.comFmailto:amitava.mukherjee@in.ibm.comFmailto:amitava.mukherjee@in.ibm.com2mailto:amitav@imit.kth.se2mailto:amitav@imit.kth.se/ 0DArialngsRomantt- 0DTimes New Romantt- 0 DWingdingsRomantt- 0@ .  @n?" dd@  @@`` TLT4    4"$$Yo( ('ib$Gv~烝+ib$Cmd@7XaL4\U "$f2䘳lI%b$]xT5WxG* n-"$  $RW.œ 8"$gvAdDZ DB 0AA@ʚ;ʚ;g4?d?d: 0wppp@ <4dddd@w 0t- 0___PPT10 d___PPT9F>&?  %.Constant Congestion Window approach for TCP  effect on Fairness  B@ (Anup K Ghosh IBM Global Services India, Salt Lake, Calcutta 700 091, India anup.ghosh@in.ibm.com Sudipto Das Dept of CSE, Jadavpur University, Calcutta 700 032, India Rajesh Roy Dept of CSE, Jadavpur University, Calcutta 700 032, India Amitava Mukherjee Royal Institute of Technology, School of Electrical Engineering, Stockholm 10044, Sweden amitava.mukherjee@in.ibm.com or amitava@imit.kth.se (Author for Correspondence) = : :> Z 0J` 0Wt 0x   Introduction  TCP performance (in terms of goodput) degrades in networks which are prone to wireless losses. Previous research has shown that End-to-End (E2E) approach towards solving this problem is the most versatile and efficient. We propose an E2E solution wherein we make the TCP Congestion Window (cwnd) a constant throughout the lifetime of the connection. The congestion window for a sender is set to a value which is optimal for a given network scenario . A cwnd for a given TCP Sender is optimal if it effectively uses its fair share of the bottleneck and the performance of the system (a number of connections sharing the bottleneck) reaches a maximum. The optimal cwnd is calculated for a given scenario and the Sender cwnd is set accordingly._PP}PPPPePPPP\P_}e\  Analytical Approach  We measure the network load by average queue length over fixed intervals of some appropriate length, and Li is the load at instant i, For a congested network we have Li = N + Li-1 (1) where N (a constant) accounts for the average arrival rate of the new traffic, and Li-1 accounts for the traffic left from the last time interval. The term Li-1 arises when the sender is sending at a rate which is greater that its fair share leading to a fraction of packets from the previous round remaining in the network when the packets from the next round arrives in the network. If the sender is sending at a rate that utilizes its fair share, the term Li-1 vanishes; equation (1) thereby reduces to Li = N (2) which is a constant, and this forms the basis for use of a constant congestion window.~PPiPPj ?   U F + - "Y1Simulation Scenario and Configuration Parameters.22Wired  cum  wireless Inter-network A performance comparison is made with the TCP Reno sources. Intermediate node buffer capacity is always set equal to the bandwidth delay product for the bottleneck link based on literature studied. The traffic source used is FTP with infinite data to send. Packet size is set to 1000 bytes (1040 bytes with headers) in all experiments. The wireless subnet is error prone, susceptible to constant error rates. Conventional TCP Sink which responds with an ACK for every packet received. No congestion or error in the ACK path. Simulations have been carried out for a period of 450 seconds with the TCP senders transmitting data for the entire period of simulation. A Two Ray Ground propagation model is used with an Omni-directional antenna. Wireless links using 802.11 MAC with 1 Mbps bandwidth All simulations done with ns-2. aZaSimulation ResultsCan be sub-divided into Three Major Categories. Optimal Window Determination. Comparison with TCP Reno in terms of Goodput. Fairness in scenarios where multiple sources share a common bottleneck.L0v0$Optimal Window Determination .The modified TCP Senders use constant Congestion Window (cwnd) throughout its Lifetime. Determination of a cwnd optimal for a given scenario becomes optimal. Results show that given a network scenario, there exists a value of cwnd for which the system performance reaches a maximum. It is the optimal cwnd value for that network scenario. The cwnd of the TCP senders is set to the value as obtained above. In the figures below, Goodput is the total number of segments received by the TCP Sink minus the retransmitted segments. Cwnd is expressed in segments..PP ,Comparison with TCP Reno in terms of Goodput-- Simulations show a 8-10% increase in goodput in comparison to TCP Reno in scenarios with wireless errors. In scenarios without wireless losses, performance of Modified TCP is comparable to that of TCP Reno. Goodput enhancement upto 17-18% in scenarios with RED queues at bottleneck routers. Enhancement with RED queues is primarily due to the dropping of more packets of the aggressive sources and hence effective sharing of bottleneck amongst the sendersP FFairness in scenarios where multiple sources share a common bottleneckGG Effective sharing of the bottleneck bandwidth when used by multiple source destination pairs is crucial for good overall system performance. Fairness might suffer in situations where a sender over utilizes its fair share of the bottleneck bandwidth. By setting cwnd of the TCP senders to the optimal value determined, all TCP Senders get an appropriate share in bottleneck bandwidth. Z FFairness in scenarios where multiple sources share a common bottleneckGG sBetter fairness obtained when the queue at the bottleneck link is replaced by RED (Random Early Detection) queues. Unfair sharing of bottleneck occurs when packets from some sources tend to occupy the bottleneck link and buffer resulting in dropping of packets from other senders. Introduction of AQM techniques like RED enhances fairness. This is because more packets of the misbehaving sources are dropped from the intermediate buffer before it is full providing buffer space for packets from other sources. Other sources (starved of bottleneck) get an opportunity to send their packets as the misbehaving sources are curbed.tPt Conclusions  %A Constant Congestion window for a TCP Sender during its entire lifetime has been proposed. Simulations show a 8-10% increase in goodput as compared to TCP Reno in scenarios with wireless error. Introduction of Active Queue Management (AQM) Techniques results in 17-18% increase in goodput compared to TCP Reno. An almost fair sharing of the bottleneck capacity achieved by setting Congestion Window to a value optimal for a given Network Scenario. Fairness enhanced by introduction of AQM techniques like RED or CHOKe at the bottleneck routers.~\PPgPPuPPPPaP& Future Enhancements In this paper, we propose a Constant cwnd for TCP Senders, but the optimal cwnd for a given network scenario is determined statically. We aim to devise a strategy by which this determination can be done dynamically. The optimal cwnd is determined for a given network scenario which is kept static throughout the lifetime of the connection. But real networks are dynamic in nature and a connection s fair share changes during its lifetime. We aim to introduce a trigger (like fractional change in rtt) that would signal such changes (usually after a large change has occurred) in fair share of the connection. We also aim to investigate the effect of other AQM techniques like CHOKe on the performance of these Modified TCP Sender.JPPPPzP2   0` 33` Sf3f` 33g` f` www3PP` ZXdbmo` \ғ3y`Ӣ` 3f3ff` 3f3FKf` hk]wwwfܹ` ff>>\`Y{ff` R>&- {p_/̴>?" dd@,|?" dd@   " @ ` n?" dd@   @@``PR    @ ` ` p>> $(    6l  `}  T Click to edit Master title style! !  0X  `  RClick to edit Master text styles Second level Third level Fourth level Fifth level!     S  0$ ^ `  X*  0 ^   Z*  0 ^ `  Z*H  0޽h ? 3380___PPT10.z`a Default Design} 0  $(  r  S }  r  S (~0`    H  0޽h ? 33___PPT10i.z@Mk+D=' e= @B +}  0 @P$(  Pr P S P `   r P S `P`  H P 0޽h ? 33___PPT10i.e+D=' e= @B +}  0 PT$(  Tr T S \: `}  : r T S  : ` : H T 0޽h ? 33___PPT10i.6+D=' e= @B +A  0 XP@(  x  c $h/:    : x  c $:@  :   PA C"  :H  0޽h ? 33___PPT10i.(+D=' e= @B +}  0 `$(  r  S ;:@  : r  S 0:P  : H  0޽h ? 33___PPT10i.pյ+D=' e= @B +Q  0 h`p (  r  S r: `p  : r  S s: `p : b  C AC"@@ ,b   C AC"p  H  0޽h ? 33___PPT10i.p+D=' e= @B +  0  ,(  , ,  B$|:"`   : r , S |:p ` : 4 , # A  z4  , # Ap L  , 0P P : 2 2  , H:"`P pComparing the TCP Reno and Modified TCP Goodput under various error scenarios and using different AQM techniques q(2p H , 0޽h ? 33___PPT10i..+D=' e= @B +  0  @(  @r @ S : `}  : x @ c $ԧ:   :  @ 0X: p 0  ~BPackets transmitted when the intermediate router queue is DropTail C(2B  @ BA "`@   :  @ 0|p@  ZThe smaller the standard deviation of the Packets transmitted, the better is the fairness.[0P[H @ 0޽h ? 33___PPT10i.^:+D=' e= @B +  0  H(  Hr H S 0: `}  : x H c $:   :  H BA P   : H 0:`   yEPackets transmitted when the intermediate router queue is a RED queueF(2F  H 0l: @  ZThe smaller the standard deviation of the Packets transmitted, the better is the fairness.[0P[H H 0޽h ? 33___PPT10i.phڧ+D=' e= @B +}  0 0 $(   r  S : `  : r  S T]:0 `S : H  0޽h ? 33___PPT10i.{pN+D=' e= @B +}  0 <$(  <r < S $.e `p  e r < S e` ` e H < 0޽h ? 33___PPT10i.pe(+D=' e= @B +r86@s`VHK$N}QVb[LCE}e 1Oh+'0x `h    ,4DConstant Congestion Window approach for TCP effect on Fairness SudiptoAdministrator5Microsoft PowerPoint@P::%@F|@ nsG<o   g  y--$ww--'@Arial-. 02 Constant Congestion Window . "System(7-@Arial-. !2 approach for TCP s.-@Arial-.  2 O.-@Arial-. "2 Ueffect on Fairness.-@Arial-.  2 0DAnup.-@Arial-.  2 0NK .-@Arial-. 2 0RGhoshv.-@Arial-. c2 5&=IBM Global Services India, Salt Lake, Calcutta 700 091, Indian.-@Arial-. '2 9?anup.ghosh@in.ibm.com.--9?9b-@Arial-. 2 >F Sudipto Das.-@Arial-. ]2 C(9Dept of CSE, Jadavpur University, Calcutta 700 032, Indian.-@Arial-. 2 HF Rajesh Roy.-@Arial-. ]2 L(9Dept of CSE, Jadavpur University, Calcutta 700 032, Indian.-@Arial-. 2 Q@Amitava.-@Arial-. 2 QP Mukherjeey.-@Arial-. g2 V@Royal Institute of Technology, School of Electrical Engineering,.-@Arial-. 2 Vu Stockholm .-@Arial-. 2 ZE 10044, Sweden.-@Arial-. 12 ^)amitava.mukherjee@in.ibm.com.--^)^V-@Arial-.  2 ^Wor .-@Arial-. $2 ^[amitava@imit.kth.seo.--^[^w-@Arial-. 02 b<(Author for Correspondence).-՜.+,D՜.+,     (On-screen Show Personalٱ_  ArialTimes New Roman WingdingsDefault DesignBConstant Congestion Window approach for TCP effect on Fairness IntroductionAnalytical Approach 2Simulation Scenario and Configuration Parameters.Simulation ResultsOptimal Window Determination-Comparison with TCP Reno in terms of GoodputGFairness in scenarios where multiple sources share a common bottleneckGFairness in scenarios where multiple sources share a common bottleneck ConclusionsFuture Enhancements  Fonts UsedDesign Template Slide Titles  8@ _PID_HLINKSAXmailto:anup.ghosh@in.ibm.com$mailto:amitava.mukherjee@in.ibm.commailto:amitav@imit.kth.se%_e 0AdministratorAdministrator  !"#$%'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXZ[\]^_`abdefghijlmnopqruRoot EntrydO)PicturesKCurrent UserkSummaryInformation(YPowerPoint Document(&eDocumentSummaryInformation8c