无线网络教学资料rrch00wirelesstcpcuiyong

1、cui yongcs 80240333wireless tcpcs 80240333 instructor: cui yong1 1cui yongcs 80240333outlinentraditional tcpnsolutions qsplit approach based solutionsqlink layer solutionsqend to end solutionsnimpact of mobility2 2cui yongcs 80240333outlinentraditional tcpnsolutions qsplit approach based solutionsql

2、ink layer solutionsqend to end solutionsnimpact of mobility3 3cui yongcs 80240333review of tcp/ipnipqpackets may be delivered out-of-orderqpackets may be lostqpackets may be duplicated 4 4nitin h. vaidya infocom06cui yongcs 80240333review of tcp/ip (contd)ntcpqconnection-orientedqreliable deliveryac

3、hieved by means of retransmissions if necessaryqcongestion avoidance and controlqend-to-end semanticsacknowledgements sent to tcp sender confirm delivery of data received by tcp receiverack for data sent only after data has reached receiver5 5cui yongcs 802403336slow startcongestionavoidanceslow sta

4、rt thresholdexample assumes that acks are not delayedreview of tcp/ip (contd)nwindow based flow controlqcongestion window size bounds the amount of data that can be sent per round-trip timeqthroughput = 3) dupacks come backqfast recovery follows fast retransmitnfast recoveryqssthresh = min(cwnd, rec

5、eivers advertised window)/2 (at least 2 mss)qretransmit the missing segment (fast retransmit)qcwnd = ssthresh + number of dupacksqwhen a new ack comes: cwnd = ssthreshenter congestion avoidance9cui yongcs 80240333congestion control10window size is reduced in half after fast retransmit and fast recov

6、ery receivers advertised windowafter fast recoverycui yongcs 80240333tcp problems in wireless networkntcp assumptionqpacket loss is due to congestionqneed to reduce the transmitting rate nbut in wireless networkqpacket loss may be caused byhigh ber, ber of a wireless channel is 10-3 - 10-5, while th

7、at of wired link is 10-8 - 10-10 or betterunstable channeluser mobilityqneed fast retransmission1111cui yongcs 80240333impact of transmission errorsnrandom errors may cause fast retransmitqretransmission of lost packetqreduction in congestion windownburst errors may cause timeoutsqtimeout results in

8、 slow start qslow start reduces congestion window to 1 mss, reducing throughputnrandom errors may also cause timeoutqmultiple packet losses in a window can result in timeout when using tcp-reno (and to a lesser extent when using sack)12cui yongcs 80240333impact of transmission errorsntcp cannot dist

9、inguish between packet losses due to congestion and transmission errorsnreducing congestion window in response to errors is unnecessaryqreduction in congestion window reduces the throughput: throughput = w / rttqideally: w = delay * bandwidthqwith loss rate p: w ( connection between mn and bs is los

10、t for small intervals of time, this leads to time out at sender; slow start at senderdata buffered by ap may be too largenmtcp solve this problem with the help of supervisory host (sh)3030remains one ack to fhz. haas icc97cui yongcs 80240333mtcp (contd)nsupervisory host qthe node in the wired networ

11、k that controls a number of apsqno caching, no retransmissionqremains one ack to fhqmonitors all packets, if disconnection detectedset sender window size to 0sender automatically goes into persistent mode (the state of the sender will not change no matter how long the receiver is disconnected)qold o

12、r new sh reopen the window31cui yongcs 80240333mtcp (contd)nadvantagesqmaintains e2e semantics (even though tcp connection is split at the supervisory host)qsupports disconnectionqno buffer forwardingqavoids slow start at the senderndisadvantagesqloss on wireless link propagated into fixed networkqa

13、dapted tcp on wireless link3232cui yongcs 80240333split-based approach : classificationnhides transmission errors from sendernprimary responsibility at base stationnif specialized transport protocol used on wireless, then wireless host also needs modification33cui yongcs 80240333split-based approach

14、 : advantagesnbs-mh connection can be optimized independent of fh-bs connectionqdifferent flow / error control on the two connectionsnlocal recovery of errorsqfaster recovery due to relatively shorter rtt on wireless link ngood performance achievable using appropriate bs-mh protocolqstandard tcp on

15、bs-mh performs poorly when multiple packet losses occur per window (selective acks improve performance for such cases)34cui yongcs 80240333split-based approach : disadvantagesnend-to-end semantics violatedqack may be delivered to sender, before data delivered to the receiverqmay not be a problem for

16、 applications that do not rely on tcp for the end-to-end semantics35fhmhbs403937383640cui yongcs 80240333nbs retains hard statebs failure can result in loss of data (unreliability)qif bs fails, packet 40 will be lost qbecause it is ackd to sender, the sender does not buffer 4036fhmhbs403937383640spl

17、it-based approach : disadvantagescui yongcs 80240333split-based approach : disadvantagesnbs retains hard statehand-off latency increases due to state transferqdata that has been ackd to sender, must be moved to new base station37fhmhbs403937383640mhnew base stationhand-off4039cui yongcs 80240333spli

18、t-based approach : disadvantagesnbuffer space needed at bs for each tcp connectionqbs buffers tend to get full, when wireless link slower (one window worth of data on wired connection could be stored at the base station, for each split connection)nwindow on bs-mh connection reduced in response to er

19、rorsqmay not be an issue for wireless links with small delay-bw product38cui yongcs 80240333split-based approach : disadvantagesnextra copying of data at bsqcopying from fh-bs socket buffer to bs-mh socket bufferqincreases end-to-end latencynmay not be useful if data and acks traverse different path

20、s (both do not go through the base station)qexample: data on a satellite wireless hop, acks on a dial-up channel39fhmhbsdataackcui yongcs 80240333outlinentraditional tcpnsolutions qsplit approach based solutionsqlink layer solutionsqend to end solutionsnimpact of mobility4040cui yongcs 80240333solut

21、ions of tcp over wirelesssnoop tcp41tcp over wirelesslink layer solutionssplit approachbased solutionsend-to-endsolutionstcp-unaware link layeri-tcpmobile-tcpelnwtcptcp sackttcpcui yongcs 80240333link layer solutionngoal: let link layer correct all errorsnmechanism qfec (forward error correction): r

22、edundancy is encoded into the messageused to correct small number of errorsfec incurs overhead even when errors do not occurqarq (automatic repeat request): retransmission at link layerused when fec capability is exceededretransmission overhead incurred only if errors occurqharq: fec+arq4242cui yong

23、cs 80240333link level retransmissions43wirelessphysicallinknetworktransportapplicationphysicallinknetworktransportapplicationphysicallinknetworktransportapplicationrxmttcp connectionlink layer statecui yongcs 80240333link level retransmissions: issuesnhow many times to retransmit at the link level b

24、efore giving up?qfinite bound - semi-reliable link layerqno bound - reliable link layernwhat triggers link level retransmissions?qlink layer timeout mechanismqlink level acks (negative acks, dupacks, )qother mechanisms (e.g., snoop, as discussed later)nhow much time is required for a link layer retr

25、ansmission?qsmall fraction of end-to-end tcp rttqlarge fraction/multiple of end-to-end tcp rtt44cui yongcs 80240333link level retransmissions: issuesnshould the link layer deliver packets as they arrive, or deliver them in-order?qlink layer may need to buffer packets and reorder if necessary so as t

26、o deliver packets in-ordernretransmissions can cause head-of-the-line blocking45receiver 1receiver 2qalthough link to receiver 1 may be in a bad state, the link to receiver 2 may be in a good stateqretransmissions to receiver 1 are lost, and also block a packet from being sent to receiver 2base stat

27、ioncui yongcs 80240333link level retransmissions: issuesnretransmissions can cause congestion lossesqattempting to retransmit a packet at the front of the queue, effectively reduces the available bandwidth, potentially making the queue at base station longerqif the queue gets full, packets may be lo

28、st, indicating congestion to the senderqis this desirable or not ?46base stationreceiver 1receiver 2cui yongcs 80240333link level retransmissions:an early studynthe senders retransmission timeout (rto) is a function of measured rtt (round-trip times)qlink level retransmits increase rtt, therefore, r

29、tonif errors not frequent, rto will not account for rtt variations due to link level retransmissionsqwhen errors occur, the sender may timeout & retransmit before link level retransmission is successfulqsender and link layer both retransmitqduplicate retransmissions (interference) waste wireless ban

30、dwidthqtimeouts also result in reduced congestion windowndisadvantage: not accurately model real tcp stacks47a. desimone globecom93cui yongcs 80240333link level retransmissions:a more accurate picturenfrequent errors increase rto significantly on slow wireless linksqlikelihood of interference betwee

31、n link layer and tcp retransmissions smallerqbut congestion response will be delayed due to larger rtoqwhen wireless losses do cause timeout, much time wastednlarge tcp retransmission timeout intervalsqgood for reducing interference with link level retransmitsqbad for recovery from congestion losses

32、qneed a timeout mechanism that responds appropriately for both types of losses (open problem)48h. balakrishnan sigcomm96cui yongcs 80240333link level retransmissions:in-order deliverynto avoid unnecessary fast retransmit, link layer using retransmission should attempt to deliver packets “almost in-o

33、rder”nin-order deliveryqnot all connections benefit from retransmissions or ordered delivery (audio)qneed to be able to specify requirements on a per-packet basis should the packet be retransmitted? how many times?enforce in-order delivery?qneed a standard mechanism to specify the requirements open

34、issue (ietf pilc working group)49r. ludwig sigmetrics98cui yongcs 80240333link layer schemes: summarywhen is a reliable link layer beneficial to tcp performance?nif it provides almost in-order deliveryand ntcp retransmission timeout large enough to tolerate additional delays due to link level retran

35、smits50cui yongcs 80240333link layer schemes: classificationnhide wireless losses from tcp sendernlink layer modifications needed at both ends of wireless linkqtcp need not be modified51cui yongcs 80240333snoop tcp n“transparent” extension of tcp within the bs(foreign agent)qbuffering of packets sen

36、t to the mobile nodeqlost packets on the wireless link (both directions!) will be retransmitted immediately by the mobile node or base station (foreign agent), qthe bs therefore “snoops” the packet flow and recognizes acknowledgements in both directionsqchanges of tcp only within the bs52“wired“ int

37、ernetbuffering of dataend-to-end tcp connectionlocal retransmissioncorrespondentnode (host)foreign agent (base station)mobilenode (host)snooping of acksh. balakrishnan acm95cui yongcs 80240333snoop tcp53fhmhbswirelessphysicallinknetworktransportapplicationphysicallinknetworktransportapplicationphysi

38、callinknetworktransportapplicationrxmtper tcp-connection statetcp connectionndata transfer to the mobile hostqfa buffers data until it receives ack of the mh, fa detects packet loss via duplicated acks (dupacks) or time-outqfast retransmission possible, transparent for the fixed networkndata transfe

39、r from the mobile hostqfa detects packet loss on the wireless link via sequence numbers, fa answers directly with a nack to the mhqmh can now retransmit data with only a very short delaycui yongcs 80240333snoop tcp nretains local recovery of split connection approach and link level retransmission sc

40、hemesqintegration of the mac layermac layer often has similar mechanisms to those of tcpthe mac layer can already detect duplicated packets due to retransmissions and discard them nimproves on split connectionqend-to-end semantics retainedqsoft state at base station, instead of hard statenhides wire

41、less losses from the sendernrequires modification to only bs (network-centric approach)54cui yongcs 80240333snoop tcp552 mbps wireless linkthroughput gain of snoop compared to base tcpcui yongcs 80240333snoop tcp: when beneficial?nsnoop prevents fast retransmit from sender despite transmission error

42、s, and out-of-order delivery on the wireless linknout-of-order delivery causes fast retransmit only if it results in at least 3 dupacksnif wireless link level delay-bandwidth product is less than 4 packets, a simple (tcp-unaware) link level retransmission scheme can sufficeqsince delay-bandwidth pro

43、duct is small, the retransmission scheme can deliver the lost packet without resulting in 3 dupacks from the tcp receiver56cui yongcs 80240333snoop tcp: advantagesnhigh throughput can be achievedqperformance further improved using selective acksnlocal recovery from wireless lossesnfast retransmit no

44、t triggered at sender despite out-of-order link layer deliverynend-to-end semantics retainednsoft state at base stationqloss of the soft state affects performance, but not correctness57cui yongcs 80240333snoop tcp: disadvantagesnlink layer at base station needs to be tcp-awarendoes not isolate the w

45、ireless link as good as i-tcpnnot useful if tcp headers are encrypted (ipsec)ncannot be used if tcp data and tcp acks traverse different paths (both do not go through the base station)58cui yongcs 80240333tcp-unaware link layerngoal qsimulate the behavior of the snoop tcp qwithout requiring the link

46、 layer at the bs to be tcp awarenat the bs, link layer retransmission is used to perform local error recoverynin snoop-tcp retransmissions are triggered by tcp duplicate acknowledgements, but in tcp-unaware link layer retransmissions are triggered by link level acks5959cui yongcs 80240333tcp-unaware

47、 link layer (contd)nmn reduces interference between tcp and link level retransmissionqfor the first two packets, sending dupacks immediatelyqfor further consecutive packets, dupacks are delayed for duration d6060cui yongcs 80240333interference in traditional tcp6161151410131011129dsnew ack1615141012

48、1311dsack1716151410131012dsack1010101817161510101413dsnew ackackack101010dup ack10cui yongcs 80240333tcp-unaware link layer (contd)6262151410131011129dsnew ack16151410121311dsack1716151410131012dsack101018171615101413dsnew ackackack1010dup ack is delayed dup ack is delayed cui yongcs 80240333tcp-una

49、ware link layer (contd)nadvantagesqlink layer need not be tcp aware.qworks well for small round trip times (rtts) over the wireless link.ndisadvantageqoptimal value of dupack delay is dependent on the wireless link. 6363cui yongcs 80240333outlinentraditional tcpnsolutions qsplit approach based solut

50、ionsqlink layer solutionsqend to end solutionsnimpact of mobility6464cui yongcs 80240333solutions of tcp over wirelesssnoop tcp65tcp over wirelesslink layer solutionssplit approachbased solutionsend-to-endsolutionstcp-unaware link layeri-tcpmobile-tcpelnwtcptcp sackttcpcui yongcs 80240333end to end

51、solutionsnonly end point participate in flow controlnthe receiver provides feedback of network conditionnthe sender makes decision for congestion control6666cui yongcs 80240333eln (explicit loss notification) nproblemqtcp doesnt know the exact cause of packet loss, assumes congestion lossnsolutionqm

52、ac layer can explicitly identify the reason for packet lossif handoff or non-congestion loss is detected, the tcp layer is immediately notified by the mac layer through eln the sender does not reduce the window size on receiving eln. this avoids slow start and can handle encrypted data6767cui yongcs

53、 80240333eln (contd)ndisadvantageq this protocol requires the mac layer of mn to be changed. the information conveyed by the mac layer might not always be reliable.68cui yongcs 80240333tcp sack (select ack)nproblemqack n acknowledges correct and in-sequence receipt of packets up to nqif single packe

54、ts are missing quite often a whole packet sequence beginning at the gap has to be retransmitted (go-back-n), thus wasting bandwidthnsolutionqtcp with selective ack qallows for acknowledgements of single packetsqallow sender to retransmit only missing packets6969m. mathisrfc 2018cui yongcs 80240333sa

55、ck (contd)nsack qindicates a block of data received at the receiver qthe sender will know the exact number of lost package nadvantageqmuch higher efficiencyndisadvantageqmore buffer at receiver qrequire modification at the sender and receiver sidesqsack blocks are encoded in the tcp option field, it

56、 limits the num of sack blocks one ack can carryqlimited ability for congestion avoidance7070cui yongcs 80240333wtcp-introductionnwtcp (reliable transmission control protocol for wide area wireless networks)nrecent have witnessed an explosive growth in the use of wireless wide-area networks (wwan)nv

57、ery low bandwidthnlarge round trip time and variance in round trip timeqas a consequence, rto values computed by the sender tend to be very highqlarge rto for loss recovery can result in poor tcp performancennon-congestion related packet lossnasymmetric channelnoccasional blackouts71prasun sinha, th

58、yagarajan nandagopal winet 2002cui yongcs 80240333wtcp solutionnany reliable transport protocol provide the following functions(a) connection management(b) congestion control(c) flow control(d) reliabilityqthe (a) connection management and (c) flow control in wtcp are similar to the standard tcp72cu

59、i yongcs 80240333wtcp solutions congestion controlnpurely end-to-end mechanismsqelimination the need for transport-level support from networknrate-based rather than window-based transmission controlqreceiver adaptively computes the desired transmission rate based on the characteristics of the data p

60、athnuse the ratio of the inter-packet delay at the receiver as the primary metric for rate controlqrather than using packet loss and retransmit timeouts73cui yongcs 80240333wtcp solutionsreliabilitynselective acknowledgmentsqcontains cumulative and selective acknno retransmit timeoutsqmodifies the s