Anda di halaman 1dari 6

ADACK: An Adaptive acknowledgment Scheme to improve link throughput and reliability for TCP over Wirele

Emmanuel Robert Ssebaggala


Dept. of Information Engineering and Computer Science University of Trento Trento, Italy ssegga@gmail.com
Abstract!Channel dynamic play a big role in the reliability of TCP tran mi ion over wirele network " #nlike wired network where the ending and receiving path are eparate$ in wirele tran mi ion the packet u e the ame medium" Thi ha led to the envi ion of an adaptive acknowledgement %ADACK& cheme that will reduce contention for the medium by efficiently modulating the number of acknowledgment ent from the TCP receiver leading to an improvement in link reliability and throughput" Keyword: TCP, Wireless Networks, Acknowledgments, WLAN

s"aring resources using pac(et/s'itc"ing among t"e nodes 596. It incorporates bot" connection/oriented lin(s and datagram services bet'een "osts. Its 'ide adoption is because of its ability to implement congestion avoidance and control, reliable ordered delivery -t"roug" retransmissions., and it:s maintenance of end/to/end semantics t"roug" reliance on ac(no'ledgements bet'een t"e sender and t"e receiver. Connection establis"ment is arc"ived t"roug" a t"ree/'ay "ands"a(e involving a se;uence of sync"roni<ation -S= C. and ac(no'ledgement -*C3. pac(ets. 3ey to its operation is t"e 'indo' management or credit allocation sc"eme '"ic" determines t"e number of pac(ets to send and at '"at time instants, 'it"out congesting t"e net'or(. T"is particular area "as been t"e focus of many researc" efforts 'it" most literature striving to find an opportune 'indo' management sc"eme. B. TCP Window Management Schemes Standard TC+ congestion control is c"aracteri<ed by an e>ponentially gro'ing slo' start p"ase during '"ic" t"e congestion 'indo' is increased by t"e number of segments ac(no'ledged. T"is continues until a predefined t"res"old or until an *C3 is not received for some segment ? t"is is assumed to be due to net'or( congestion. TC+ t"en reduces t"e t"res"old and enters congestion avoidance. *t t"is point, t"e 'indo' is increased by & segment for eac" round trip time. In TC+ Ta"oe, triple duplicate *C3s are treated t"e same 'ay as timeouts. Ta"oe performs @fast retransmitA to reduce t"e congestion 'indo' to one ma>imum segment si<e -BSS. and t"en resets to slo'/start. In TC+ Reno, if t"ree duplicate ac(no'ledgements -DU+*C3S. are received, t"e congestion 'indo' 'ill be "alved. T"is is follo'ed by a fast retransmit p"ase, t"en fast recovery. If an *C3 times out, slo' start is used as it is 'it" Ta"oe. TC+ e' Reno improves t"e fast recovery p"ase of TC+ Reno. Several variants of t"e TC+ protocol focusing on 'indo' management "ave been proposed. 4rat(o and +eterson in &CC9 proposed a different approac", called TC+ 8egas, '"ic" reacted to delays -i.e., variations of t"e RTT. as congestion signals 5D6.Unfortunately its incremental deployment results in unfairness and poor performance in comparison to e' Reno.

I.

TR!DUCTI!

T"e IEEE #$%.&& standards provide t"e basis for 'ireless net'or( communication '"ic" is becoming more prevalent in )ocal *rea et'or( implementations because of its reduced cost of o'ners"ip, initial installation speed, scalability and mobility. T"is "as increased t"e need to improve t"e performance and reliability of TC+ over 'ireless net'or(s. ,o'ever, t"e Transmission Control +rotocol -TC+. 'as originally designed for reliable end/to/end data transmission over 'ired net'or(s. Its application in 0ireless environments faces problems suc" as pac(et losses due to congestion, "ig" bit error rates, mobility, etc1 2urt"ermore, 'ired net'or(s "ave a t'o/ pat" medium for sending and receiving TC+ pac(ets t"erefore c"annel contention is not a big issue as is t"e case in 'ireless net'or(s '"ere t"e medium is t"e same for bot" data pac(ets and ac(no'ledgments -*C3s.. 4ecause of t"is, 'ireless net'or(s suffer more from delays and transmission errors 5%6576 from collisions and pac(et losses t"us reliability and its assurance are (ey issues. T"is paper revie's TC+, t"e #$%.&& standards and some of t"e various ac(no'ledgement/based sc"emes found in literature and discusses t"eir s"ort falls. It t"en proposes a sc"eme t"at 'ill improve c"annel reliability t"roug" adaptive variation of t"e ac(no'ledgment rate. II. )ITER*TURE RE8IE0

A. Transmission Control Protocol TC+, as first envisioned by 8int Cerf and Robert 3a"n, 'as t"e central component for an internet'or(ing protocol for

*s a 'ay to improve TC+ performance in t"e presence of lin( losses -especially in 'ireless., TC+ 0est'ood and 0est'oodE (eep an estimate of t"e available band'idt" '"ic" is used to implement an *dditive Increase *daptive Decrease -*I*D. strategy 5F65G6. T"eir problem is t"at band'idt" estimation updated once per RTT, suffers in presence of long RTT and fast c"annel dynamics. Tom 3elly proposed a Bultiplicative Increase Bultiplicative Decrease -BIBD. strategy to overcome t"e limitations of TC+ in very large 4D+ net'or(s called Scalable TC+ -S/TC+. 5#6. Unfortunately, S/TC+ does not represent a c"oice for t"e current Internet, since it induces too many fluctuations in t"e net'or( be"avior. S.2loyd proposed a ,ig"Speed TC+ -,STC+. implementation in '"ic" t"e congestion 'indo' dynamic is similar to TC+ Reno for lo' congestion 'indo' values, and approac"es S/TC+ for large congestion 'indo' values 5C6. 2*ST TC+ -proposed by S. )o' et al.. is based on t"e use of delays as congestion signalsH it "as t"e advantage of more accurate delay estimation t"an losses in large band'idt" delay product -4D+. net'or(s 5&$6.2*ST ac"ieves very "ig" lin( utili<ation but it suffers from stability problems in t"e presence of bursty reverse traffic and incremental deployment is still unclear. ). Iu et al. in %$$9 proposed 4inary Increase -4IC. TC+ based on a binary searc" tec"ni;ue, '"ere t"e sender tries to guess t"e optimal operating point and gets a binary feedbac( by t"e net'or( 5&&6. T"e increase p"ase is linear at small congestion 'indo' values and gets logarit"mic as t"e sender approac"es pat" capacity '"ile t"e decrease is still multiplicative. 4IC TC+ 'or(s 'ell under a range of operating conditionsH it is suitable to incremental deployment and does not result in oscillations. CU4IC TC+ is an en"anced version of 4inary Increase Congestion Control 5&%6. It simplifies t"e 4IC 'indo' control function and improves its TC+/ friendliness and RTT fairness as 4IC:s gro't" function is too aggressive for TC+ especially under s"ort RTT or lo' speed net'or(s 5&76. T"e 'indo' gro't" function is cubic in terms of t"e elapsed time since t"e last loss event. CU4IC provides good scalability and stability. T"e protocol (eeps t"e 'indo' gro't" rate independent of RTT (eeping t"e protocol friendly under long and s"ort RTTs. Compound TC+ integrates a scalable delay/based component into t"e standard TC+ congestion avoidance algorit"m 5&96 5&76. T"is "as a fast 'indo' increase function '"en t"e net'or( is under/utili<ed and reduces t"e sending rate '"en a congestion event is sensed. 2.Jranelli et al. proposed en"ancements to TC+ 0est'oodE by employing a 4IC/li(e logarit"mic increase in t"e congestion 'indo' 5&D65&F6. T"e result 'as TC+ )og0est'oodE, based on a )ogarit"mic Increase *daptive Decrease -)I*D. strategy. T"e en"ancements provide fast 'indo' increase for lo' 'indo' values, '"ile being accurate in approac"ing t"e available pipe capacityH t"ey also result in small sensitivity 'it" respect to RTT.

C. 802.11 Standards T"e IEEE #$%.&& 0ireless )ocal *rea et'or( -0)* . standards 'ere first adopted in &CCG and several revisions "ave follo'ed t"ereafter. T"ey specify an over/t"e/air interface bet'een a 'ireless client and access point. T"e standards differ from t"e #$%.I )* standards at t"e Bedium *ccess Control -B*C. layer. T"e B*C protocol provides t'o access met"odsK Distributed Coordination 2unction -DC2. ordinary async"ronous communication and +oint Coordination 2unction -+C2. for communication 'it" an access point. T"e DC2 is a Carrier Sense Bultiple *ccess +rotocol -CSB*. 'it" collision avoidance -CSB*LC*.. It uses @Re;uest to sendA -RTS. and @Clear to sendA -CTS. pac(ets to avoid collision in t"e medium. T"e e>c"ange of t"ese signals ensures t"at ot"er terminals defer transmission for some period of time -called t"e et'or( *llocation 8ector. until t"e medium is free 5&G6. T"e et'or( *llocation 8ector - *8. represents t"e duration t"at stations sensing RTSLCTS signals "ave to leave t"e medium in order to allo' successful delivery of bot" a data frame and t"e corresponding ac(no'ledgement. 0"en t"e traffic of a station is bac(logged, RTS messages are sent to notify t"e ot"er stations t"at it 'ants to transmit a pac(et to a specific receiver. T"e receiver t"en notifies t"e sender t"at it can transmit by sending a CTS message. o ot"er station transmits any pac(et in a period corresponding to t"e transmission time of t"e pac(et to be sent. T"e small si<e of RTSLCTS control frames ma(es t"e probability of collision during t"eir transmission lo'er t"an t"at of t"e collision of data frames, since t"ese frames are usually larger t"an t"ose involving RTSLCTS frames. T"e #$%.&& standards suffer significant performance limitations. T"eir performance depends on bot" t"roug"put and associated delays '"en CSB*LC* is employed. T"e most relevant issues areK Bandwidth: T"e total lin( capacity is s"ared by all nodes '"ic" can operate 'it"in transmission range, including "idden terminalsH since collisions dramatically decrease t"e t"roug"put, it is desirable to monitor it 5&G6 5&#6. Latenc K )atency in a 'ireless medium is greater t"an in a 'ired one. It is affected by propagation delays, t"e over"ead added by bot" p"ysical layer and lin( layer protocols and t"e retransmission policy implemented at t"e lin( layer Channel losses: 0ireless c"annels suffer from fading caused by interference 'it" ot"er sources. 0"ile t"e 4it Error Rate -4ER. varies from &$ /F to &$/# in 'ired c"annels, it varies from &$/7 and &$/& in 'ireless c"annels 5&G6 5&C6. Mo!ilit K T"e most common net'or( setting is a -4ase Station Subsystem. 4SS connected to a fi>ed net'or( via a 4ase Station -4S.. ,andoff -S'itc"ing bet'een t"ese 4Ss. re;uires t"at all t"e information associated 'it" user activities be transferred from 4S to t"e ne>t to prevent t"e termination of service provided to t"e mobile user 5&G6.

TCPK T"e problems t"at arise in t"e usage of TC+ over 'ireless net'or(s are due to t"eir lo' reliability, as 'ell as time/variant c"aracteristics suc" as fading, s"ado'ing, node mobility, "and/offs, limited available band'idt", and large RTTsH TC+ performs poorly in suc" environments 5%$6 5%&6.

conceptsH t"e result 'as D*0) 5%F6. In D*0), cumulative *C3s are encapsulated into data frames, standalone *C3s are neglected, frame loss detection is based on t"e analysis of se;uence numbers, and negative selective *C3s allo' immediate re;uest. D*0) reduces medium/busy time. D. 3lia<ovic", 2. Jranelli also proposed +ac(et Concatenation at t"e I+ )evel -+*C/I+. '"ic" en"ances performance t"roug" concatenation of I+ pac(ets -to t"e same destination. in a single data lin( pac(et 5%G6. It provides good performance for small pac(ets. !n t"e ot"er "and, transport layer solutions try to ac"ieve better performance t"roug" t"e modification of TC+ itself. T"ey "ave t'o subsetsK connection splitting and TC+ modifications. Connection splitting approac"es brea( end/to/ end TC+ connection into a connection on a 'ired route and ot"er on a 'ireless route. T"e split connection results in independent flo' control for t"e t'o parts. T"is provides more degrees of freedom in t"e optimi<ation of t"e TC+ over t"e 'ired as 'ell as on t"e 'ireless lin(. E>amples are I/TC+ 5%#6, and BET+ 5%C6. TC+ Bodifications do not greatly c"ange t"e be"avior of TC+, but only c"ange some of mec"anics used by TC+ -suc" as, ac(no'ledgement generation mec"anism.. E>amples include selective ac(no'ledgement '"ic" enables t"e receiver to inform TC+ sender about pac(ets received out of order 57$6. ,aving t"is information, t"e sender retransmits only t"ose pac(ets '"ic" "ave not reac"ed t"e receiver. *C3 tec"ni;ue performs better t"an standard TC+ in case of multiple losses 'it"in a single 'indo'. TC+ Santa Cru< is anot"er TC+ Bodification transport layer solution. It employs t"e options field of t"e TC+ "eader for t"e implementation of ne' congestion control and error recovery strategies 57&6. T"e increased amount of information about t"e TC+ flo' proves a better RTT estimation to sender. It also implements selective ac(no'ledgement. E>plicit otification Sc"emes -E S. i.e. E>plicit 4ad State otification -E4S . and E>plicit )oss otification -E) . send notification to t"e sender '"en an error occurs at t"e 4S. E4S notifications are sent '"enever t"e 4S is unsuccessful in message delivery 57G6. Upon t"e reception of suc" a message t"e TC+ sender restarts its timer in order not to go into slo' start since no congestion occurred. T"is sc"eme re;uires some minor modifications to t"e TC+ sender code. E) notifications on t"e ot"er "and ma(e t"e sender a'ay of non/congestive errors occurring in t"e net'or( 5776. T"e 4S monitors TC+ pac(ets in bot" directions. 0"en a duplicate ac(no'ledgement is sent from t"e TC+ receive, t"e 4S -"aving (no'ledge of t"e type of error. encapsulates t"e E) message by setting up t"e E)4 bin in t"e TC+ ac(no'ledgement field and t"en for'ards it to t"e sender. In E S sc"emes, t"e modified sender can c"oose "o' to react according to t"e type of loss t"at "as occurredH it could go to congestion avoidance, or c"oose some ot"er more appropriate 'ay. Unfortunately, E) doesn:t provide local retransmission. T"e last category of TC+ improvement over 'ireless is cross layer solutions. T"e brea( t"e IS!L!SI layering principles bay allo'ing interdependence and Ooint design of

T"is proposal focuses on issues in"erent to TC+:s implementation t"at compromise it '"en it is used over 'ireless lin(s. Specifically, its ac(no'ledgement sc"eme is e>ploited for better performance over 'ireless lin(s. ". Schemes to im#ro$e the #er%ormace o% TCP o$er Wireless Sc"emes to improve TC+ performance over 'ireless can be broadly categori<ed into t"ree groupsK lin( layer solutions, transport layer solutions, and cross layer solutions 5&G6. )in( layer solutions retransmit a pac(et at t"e lin( layer, if errors are detected e.g. 2or'ard Error Correction -2EC.. T"ey can be furt"er bro(en do'n into TC+/a'are and TC+/una'are solutions. E>amples of TC+/a'are lin( layer solutions are Snoop protocol and 0ireless Transmission Control +rotocol -0TC+.. 0TC+ is a pro>y based modification of TC+ t"at preserves t"e end/to/end semantics of TC+. 0TC+ is placed on a 4S bet'een a fi>ed "ost -2,. and a mobile "ost -B,. "ost. 0"en t"e B, uses its TC+ to send a segment, t"e 0TC+ at t"e 4S receives it and sends it on t"roug" t"e net'or(, '"ere it eventually reac"es t"e a'aiting "ost. T"e a'aiting "ost mig"t send an ac(no'ledgment bac( t"roug" t"e net'or(, to t"e 4S, '"ic" transmits it to t"e B,. Despite "andling some 'ireless/ related errors, 0TC+ effectively does e>actly '"at regular TC+ does. T"e t'o edge "osts arenMt even a'are t"at t"e 0TC+ e>ists 5%76 5%96. T"e Snoop protocol introduces a Snoop module -or 4ase Station. bet'een t"e B, and 2,. T"e connection bet'een t"e 2, and 4S is fi>ed '"ile t"e connection bet'een t"e B, and 4S is 'ireless 5%%6. T"e snoop agent maintains a cac"e of TC+ pac(ets sent from t"e 2, t"at "ave not yet been ac(no'ledged by t"e B,. In case t"e pac(et from 2, is a duplicate, t"e cac"e is only updated 'it"out furt"er pac(et propagation to t"e B,. Retransmission of t"e cac"ed pac(ets is based on t"e reception of duplicate ac(no'ledgements from B,. Upon successful retransmission, duplicate ac(no'ledgements are dropped at t"e 4S to avoid application of t"e fast retransmission and congestion control procedures on t"e 'ired "ost. TC+/una'are lin( en"ance lin( layer performance by implementing error correction tec"ni;ues suc" as for'ard error correction. T"ey also implement various *utomatic Repeat Re;uest -*RN. sc"emes for t"e retransmission of loss pac(ets on t"e lin( layer e.g. *IRB*I), TU)I+, Delayed DU+*C3s -DD*., and D*0I) 5&G6. T"e DD* protocol implements lin( layer retransmission at t"e 4S and reduces interference bet'een TC+ and lin( layer retransmissions at t"e B, by delaying DU+*C3s 5%D6 5&G6. DU+*C3s delay is intended to give lin( layer enoug" retransmit time to succeed. It "as t"e disadvantage of delaying t"e recovery from losses. D. 3lia<ovic" and 2. Jranelli combined DD* 'it" Service Specific Connection !riented +rotocol -SSC!+. based

protocols across different layers. T"is is interesting approac" 'it" fe' solutions availableH some of t"ese are t"e Interlayer Collaboration +rotocol -I)C/TC+. and )in( )ayer *RN E>ploitation TC+ -))E/TC+. 5&G6. I)C/TC+ is designed for t"e performance improvement of TC+ in 'ireless environments, involving fre;uent and long disconnections as 'ell as in t"e presence of mobility 5796. T"e main concept be"ind it is t"e introduction of a State Banager -SB. in parallel 'it" t"e protocol stac( for gat"ering information from TC+, I+, and )in(L+"ysical layers. If necessary, t"e collected information is provided upon a re;uest from t"e ot"er layers. Eac" )in(L+"ysical or I+ layer periodically reports to SB about its state. T"e differentiation is done by t'o levels -!3 or !T !3. depending on "o' t"e current level can "andle pac(et delivery. ))E/TC+ aims at optimi<ing t"e usage of t"e *RN sc"emes involving t"e operation of TC+ over #$%.&& 'ireless lin(s 57D6. It e>ploits information about t"e successful pac(et delivery provided by t"e lin( layer in order to en"ancement performance. T"e transmission of stand/alone TC+ *C3s is neglected and instead an *RN snoop agent is implemented. T"e snoop agent able to interact 'it" t"e B*C entity for notification delivery and t"e TC+ entity for delivering TC+ *C3s and buffering sent pac(ets.

2igure &K Simplistic *D*C3 !peration 0"en t"e positive *C3 -+*C3. is received, t"e time/out counter is updated 'it" consideration of t"e time it 'ill ta(e t"e pac(ets to reac" t"e receiver, be processed, and t"e reception of t"e +*C3 for t"e 'indo'.

III.

DESCRI+TI! !2 +R!+!SED BET,!D

A. &eneral "escri#tion T"e proposed met"od ? *daptive *c(no'ledgement -*D*C3. ? is a slig"t modification of D. 3lia<ovic" and 2. Jranelli:s D*0) protocol 5%F6. T"e (ey concept in t"is approac" is to use t"e statistics of pac(ets received during t"e current congestion 'indo' to develop a model for t"e arrival time, and inter/arrival time '"ic" are c"aracteristic of t"e current state of t"e net'or(. 0e t"en use t"is information as prior (no'ledge for processing t"e pac(ets in t"e ne>t congestion 'indo'. *ssuming a Jaussian model for t"e arrival times, at t"e end of eac" 'indo' se;uence, a positive *C3 -+*C3. 'ill be sent bac( to t"e sender 'it" t"e mean and variance of t"e arrival rates and possibly 'it" t"e for'ard transmission time '"ic" is based on t"e first pac(et of eac" 'indo'. T"e sender 'ill use t"ese statistics to appropriately adOacent its timeout counter for t"e ne>t 'indo' of pac(ets it sends. 2igure & is a simplistic reali<ation of t"is approac".

B. Com#'ting the c'rrent lin( statistics *s stated earlier, t"e statistics generated in t"e current 'indo' are used for t"e second 'indo'. *ssuming a Jaussian distribution of t"e arrivals and inter/arrival times, t"e mean and variance for t"e current 'indo' is computed. Running values s"ould be maintained in order to avoid dealing 'it" t"e storage of all t"e pac(ets of eac" 'indo'H '"enever a ne' pac(et arrives t"e mean and variance are updated. T"e first pac(et of eac" 'indo' carries information about t"e si<e of t"e 'indo'. To build t"e mat"ematical tools necessary for t"is approac", 'e assume pac(ets of t"e current 'indo' of si<e 0 "ave been received. Jiven a pac(et pi, -'"ere i is t"e position in t"e 'ido' and can be determined from t"e se;uence number., arriving at time t i t"e inter/arrival time is given by =1 . (1)

0"ere ti is t"e time of arrival for t"e i/t" pac(et and a i is t"e inter/arrival time bet'een t"e i/t" pac(et and t"e previous pac(et. T"e running average for t"e inter/arrival time can be computed from =1 + (1) / (2)

0"ere ui is t"e mean inter/arrival time '"en t"e i/t" pac(et is received, and i is t"e inde> of t"e current pac(et in t"e 'indo'. T"e variance is computed from = 1 + () 2. 2= /(1) (3) (4)

treated as out/of/order instead of being treated as lost. T"is e;uation ensures t"at a missing pac(et in t"e first "alf of t"e 'indo' is treated as out/of/order. T"is allo's enoug" time for out/of/order pac(ets to be received '"ile still responding to pac(et losses in time. T"is eliminates unnecessary sending of *C3 pac(ets t"at 'ould end up congesting t"e lin(. ". Timeo'ts and B'%%er Timeouts are computed by using t"e statistics of t"e arrivals at t"e receiver as developed earlier. T"is information, toget"er 'it" an appropriate model can be used to calculate an optimum value for t"e time out counters '"ic" allo's incorporation of lin( dynamics. 4uffering is "andled t"e same 'as as in D*0). C! C)USI! T"e nomenclature for t"e proposed met"od is motivated by t"e fact t"at one *C3 is sent for eac" 'indo' of received pac(ets. In effect, t"is produces a variable ac(no'ledgement rate t"at depends on t"e si<e of t"e 'indo'H as t"e 'indo' increases, t"e rate reduces and vice versa. In essence t"e ac(no'ledgement rate adapts to t"e 'indo' si<e and "ence t"e c"annel dynamics. In conclusion, t"is approac" promises to offer t'o main advantages over D*0)H it effectively reduces t"e number of ac(no'ledgements "ence reducing collisions and allo's for t"e protocol to set appropriate time/out values. T"is is ac"ieved by incorporating c"annel dynamics t"roug" using a simple statistical model for t"e arrival and inter/arrival distributions. RE2ERE
5&6 CES

T"is is also a running computation. To account for errors and out of order pac(ets, t"e inde> i can be ta(en as t"e position of arrival at t"e receiver. C. )rror *andling with A"AC+. In case of an error, *D*C3 employs t"e same mec"anism used by D*0) to provide fast error recovery. * *C3 is sent re;uesting t"e retransmission of t"e missing pac(et. *fter successful retransmission of t"e lost frames, t"e transmitter:s data/Po' continues. If t"e sender does not receive a positive ac(no'ledgement for t"e transmitted data 'it"in +!)) timeout time, it 'ill poll t"e receiver t"roug" transmission of +!)) frame.

2igure % Error ,andling 2rom 2igure %, '"en t"e sender timeout counter e>pires before receiving a +*C3 pac(et, a +!)) pac(et is sent re;uesting retransmission of t"e +*C3. *D*C3 introduces an efficient 'ay of differentiating bet'een lost pac(ets and out/of/order pac(ets. 0 / p Q 0L% E 7 (5)

0"ere p is t"e pac(et position in t"e 'indo' -t"is is determined from t"e se;uence number. and 0 is t"e congestion 'indo' si<e -got from t"e first frame of t"e 'indo'.. If e;uation -D. is satisfied, a missing pac(et is

IEEE Standard for 0ireless )* Bedium *ccess Control -B*C. and +"ysical )ayer -+,=. SpeciRcations , IS!LIEC ##$%/&&KK&CCC-E., *ugust &CCC. 5%6 2abri<io Jranelli, D<mitry 3lia<ovic", and elson ). S. Da 2onseca, @+erformance )imitations of IEEE #$%.&& et'or(s and +otential En"ancements,A 576 0. Stevens. TC+LI+ illustrated volume &. *ddison 0esley ,&CCC. 596 8inton J. Cerf, Robert E. 3a"n, -Bay &CG9.. @* protocal for +ac(et eto'r( InterconnectionA 5D6 )a'rence S. 4ra(mo, )arry ). +eterson,and Sean 0. !MBalley, @TC+ 8egasK ne' tec"ni;ues for congestion detection and avoidanceA, +roceedings of t"e SIJC!BB MC9 Symposium , *ugust &CC9, pg. %9/7D 5F6 S. Bascolo, C. Casetti, B. Jerla, B. Sanadidi, R. 0ang, TC+ 0est'oodK end/to/end band'idt" estimation of efScient transport over 'ired and 'ireless net'or(s, inK +roceedings of t"e *CB Bobicom, Rome, Italy, %$$& 5G6 D. 3lia<ovic", 2. Jranelli, D. Biorandi, TC+ 0est'oodE en"ancements for "ig"/speed long/distance net'or(s, inK +roceedings of t"e IEEE ICC, Istambul, Tur(ey, %$$F. 5#6 Tom 3elly, Scalable TC+K Improving +erformance in ,ig"speed 0ide *rea et'or(s, %$$7 5C6 S.2loyd, R2C 7F9CK ,ig"Speed TC+ for )arge Congestion 0indo's, December %$$7 5&$6 David I. 0ei C"eng Tin Steven ,. )o' SanOay ,egde, 2*ST TC+K Botivation, *rc"itecture, *lgorit"ms, +erformance, December, %$$D 5&&6 )isong Iu, 3"aled ,arfous", and InOong R"eeK @4inary Increase Congestion Control for 2ast, )ong Distance et'or(sA, %$$7 5&%6 InOong R"ee, and )isong IuK @CU4ICK * e' TC+/2riendly ,ig"/ Speed TC+ 8ariantA, Sangtae ,a, InOong R"ee, )isong Iu.

5&76 ,abibulla" Tamal, 3iran Sultan, +erformance *nalysis of TC+ Congestion Control *lgorit"ms, in International Tournal of Computers and Communications, %$$# 5&96 3un Tan, Tingmin Song, Nian U"ang, Burari Srid"aran, A* Compound TC+ *pproac" for ,ig"/speed and )ong Distance eto'r(sA,&CCD 5&D6 D. 3lia<ovic", 2. Jranelli, and D. Biorandi, VTC+ 0est'oodE En"ancements for ,ig"/Speed )ong/Distance et'or(s,@ ICC,%$$F 5&F6 D<mitry 3lia<ovic", 2abri<io Jranelli , and Daniele Biorandi , @)ogarit"mic 'indo' increase for TC+ 0est'oodE for improvement in "ig" speed, long distance net'or(sA 5&G6 2abri<io Jranelli, D<mitry 3lia<ovic", and elson ). S. Da 2onsenca, @+erfomance )imitations of IEEE #$%.&& et'or(s and +otential En"ancementsA, 5&#6 ). C"eng and I. Barsic , @)ig"t'eig"t Bodels for +rediction of 0ireless )in( Dynamics in 0irelessLBobile )ocal *rea et'or(sA , in +ro ceedings of t"e IEEE %$$% SarnoW Symp osium on *dvances in 0ired and 0ireless Communications, pp. C#/&$&, T, US*, Barc" &7, %$$%. 5&C6 3. +enti(ousis , TC+ in 0ired/Cum/0ireless , @IEEE Communications SurveysA, %$$$ 5%$6 3. 2all and S. 2loyd , Simulation/based Comparisons of Ta"oe, Reno, and S*C3 TC+ , Computer Communication Revie' &CCF 5%&6 B. C"inta and S. ,elal , I)C/TC+K *n Interlayer Col laboration +rotocol for TC+ +erfor/mance Improvement in Bobile and 0ireless Environments , +roceedings of t"e Trird IEEE 0ireless Communications and et'or(ing Conference -0C C., e' !rleans, )ouisiana,Barc" %$$7. 5%%6 4ela(ris"nan, ,., Ses"an, S., *mir, E., 3at<, R., @Improving TC+LI+ +erformance over 0ireless et'or(s,A B!4IC!B, &CCD. 5%76 Ratnam 3aruna"aran and Ibra"im Batta, 0TC+K *n Efficient Bec"anism for Improving 0ireless *ccess to TC+ Services, %$$% 5%96 Sin"a +rasun et al., 0TC+K * Reliable Transport +rotocol for 0ireless 0ide/*rea et'or(s, &CCC

5%D6 8aidya, .,., Be"ta, B., +er(ins, C., Bontenegro, J., @Delayed Duplicate *c(no'ledgementsK * TC+/una'are *pproac" to Improve +erformance of TC+ over 0ireless,A Tec".Rep. CC/$$7, Computer Science Department, Te>as *XYB University, 2ebruary &CCC 5%F6 D. 3lia<ovic", 2. Jranelli, @* Delayed/*C3 Sc"eme for B*C/)evel +erformance En"ancement of 0ireless )* s,A ICT:$9, 2ortale<a -4ra<il., *ugust &/G, %$$9 5%G6 D. 3lia<ovic", 2. Jranelli, @+ac(et Concatenation 'it" NoS support for 0ireless )ocal *rea et'or(s,A Tec". Report ZDIT/$9/$%D, *pril %$$9. 5%#6 4a(re, *., 4adrinat", 4.R., @I/TC+K Indirect TC+ for mobile "osts,A Distributed Computing Systems, +roceedings of t"e &D/t" International Conference, pages &CF / &97, Tune &CCD 5%C6 3uang/=e" 0angH Tripat"i, S.3., @Bobile/end transport protocolK an alternative to TC+LI+ over 'ireless lin(s,A &Gt" *nnual Toint Conf. of t"e IEEE Computer and Comms Societies, 8ol. 7, %C Barc"/% *pril &CC# +agesK&$9F ? &$D7 57$6 rfc %$&# , "ttpKLL'''.ietf.orgLrfcLrfc%$&#.t>t, !ctober, &CCF 57&6 C. +arsa and T.T. Jarcia/)una/*ceves, @Improving TC+ Congestion Control over Internets 'it" ,eterogeneous Transmission Bedia,A +roceedings of IEEE IC + MCC, Toronto, !ctober &CCC 57%6 Ref.K 4.S. 4a(s"i, +. 3ris"na, .,. 8aidya, and D.3. +rad"an, @Improving performance of TC+ over 'ireless net'or(s,A +roceedings of t"e &Gt" International Conference on Distributed Computing Systems &CCG, 4altimore, Bay &CCG 5776 ,. 4ala(ris"nan and R. ,. 3at<, @E>plicit )oss otification and 'ireless 'eb performance,A +roc. IEEE Jlobecom Internet Bini/Conference, Sydney, *ustralia, ov. &CC# 5796 C"inta, B., ,elal, S., @I)C/TC+K *n Interlayer Collaboration +rotocol for TC+ +erformance Improvement in Bobile and 0ireless Environments,A +roc. 0C C, e' !rleans, )ouisiana, Barc" %$$7 57D6 D. 3lia<ovic", 2. Jranelli, @* Cross/layer Sc"eme for TC+ +erformance Improvement in 0ireless )* s,A J)!4EC!B%$$9

Anda mungkin juga menyukai