Peer-to-Peer Protocols and Data Link Layer. Chapter 5 from Communication Networks Leon-Gracia and Widjaja

Transcription

1 Peer-to-Peer Protocols and Data Link Layer Chapter 5 from Communication Networks Leon-Gracia and Widjaja

2 Peer-to-Peer Protocols At each layer two (or more) entities execute These are peer processes For process at layer n Process at layer n+1 requests transfer of an SDU Process at layer n constructs a PDU Passes it to process n-1 Eventually process at layer n at the other side is notified Process at layer n opens the PDU received Passes the original SDU to layer n+1

3 Peer-to-Peer Protocols n+1 n+1 n n n-1 n-1

4 Service Models Connection Oriented Connectionless Transfer capability Size of blocks (especially connectionless) Streams (arbitrary size) Constant/variable rate QoS Level of reliability Max delay Real-time, best-effort,...

5 Types of Services Arbitrary message size or structure Sequencing Reliability Timing Flow control Multiplexing Privacy, integrity, authentication

6 Segmentation and Blocking If an SDU is too big we segment into smaller pieces How to reassemble Many small SDUs can be combined to large blocks With error detection, re-transmission and sequencing we can achieve reliable connection from unreliable channels

7 Finite Buffering If we have small buffers at the receiving end, packages can be lost Today the problem is due to traffic congestion in intermediate nodes Cheap solution: drop packages and rely on ARQ Better solution provide flow control

8 Multiplexing and Security We almost always share a connection Security threats can arise DoS Trojan horses Man in the middle Eavesdropping

9 End-to-End vs Hop-by-Hop Transport Transport Network Network Network Network Data link Data link Data link Data link Physical Physical Physical Physical

10 End-to-End vs Hop-by-Hop In connectionless layers, if they arrive at all Two packets may arrive out of order A packet may arrive double This can happen Follow different paths Some intermediate node lost a packet and was later retransmitted A packet was thought lost but finally arrived

11 Two approaches Error control at every hop Use CRC, and timeouts Transmit ACK or NAK Error control at the ends Use similar techniques Trade-off: Responsiveness Overhead

12 Examples TCP Uses end-to-end HDLC (High-level Data Link Control) Uses hop-by-hop

13 Automatic Repeat request ARQ for short Combine error detection, retransmission We examine three basic types First: assume transfer of information in one direction (control signals can flow in both) We call layer n+1 PDUs packets We call layer n PDUs frames

14 ARQ The header should contain the usual stuff plus CRC bits Type: I-Frames or Control-Frames Have a time-out mechanism to detect lost packets First assume wirelike transfer In order Like ATM

15 Stop and Wait The simplest protocol: Send a frame with CRC etc Wait until we receive an ACK If we receive an ACK continue If we receive a NAK or timeout re-transmit Could not be simpler to implement

16 Stop and Wait Timeout Fr-1 ACK-1 Fr-2 Fr-2 ACK-2 Timeout Fr-1 ACK-1 Fr-2 ACK-2 Fr-2 ACK-2

17 Stop and Wait Timeout Fr-1 ACK-1 Fr-2 ACK-2 Fr-2 ACK-2 Timeout Fr-1 ACK-1 Fr-2 ACK-2 Fr-2 ACK-2

18 Problem We cannot distinguish between the two situations Sender does not know if receiver got the message or not Lost packets and delayed ACKs can be a problem

19 Solution Add sequencing numbers How about if we have long sequences We cycle back to zero What is the minimum number of bits We can do it with just one bit

20 The Protocol Transmitter can be either READY or WAITING READY: wait for higher level to request service WAITING: wait for an ACK or TIMEOUT When in READY and receives a service request, creates a frame including the S last and goes into WAITING When in WAITING and receives TIMEOUT: retransmits ACK: if CRC and S last correct, increment S last and go to READY

21 The Protocol The receiver If a frame arrives and S last =R next, increment R next and send ACK with R next. If a frame arrives and S last <>R next, do not increment R next and send ACK with old R next.

22 Performance An important quantity is the delay-bandwidth product It is the number of bits that could be transmitted in the time it takes to send a frame and receive its ACK If it is much bigger than the frame the system is inefficient

23 Performance

24 Performance t 0 t prpg t f t proc t proc t ACK

25 Performance η 0 = n f n o t 0 R Delay-Bandwidth product

26 Performance η 0 = n f n o t 0 R t f = n f R t prpg = D c t ack = n ack R t o =2 t prpg +t f +t ack +t proc t o R=2 D R c +n f +n ack +t proc R

27 Performance If the distance is small and the overhead/ack is small, works well Very simple Not practical if distance is large or overhead/ack substantial

28 Go-Back-N ARQ We can do better We can keep transmitting Have up to W s un-acknowledged If the ACK for the oldest un-acknowledged timesout re-transmit last W s If an ACK received and is OK consider that frame and all before it acknowledged We can have up to W s frames in the pipeline This is a sliding window technique

29 Go-Back-N ARQ timeout

30 Go-Back-N ARQ Frames Received And acked S last S recent S last W s 1 Frames Received And acked R next S last S last S recent... S last +W s 1

31 The Protocol (Transmitter) While buffers available, accept service requests from layer above, package the packet into a frame, send it and increment S recent. Start timer. If a correct ACK arrives mark all the older frames as transmitted successfully. Slide window If a timeout occurs, retransmit all frames and reset timers

32 The protocol (Receiver) Same as stop-n-wait When frame arrives Check for errors Check if seq. Number is same as Rnext. Increment Rnext Send ACK

33 Maximum Window Size Should be 2 m -1 If it was 2 m then the receiver would not know if it was next frame or his ACKs were lost and it is retransmission

34 Bidirectional Links Piggyback control frames onto I-frames If frame arrives correctly send ACK with next departing I-frame If no I-frame currently wait a short period If nothing arrives send a control frame Adjust time-out period to account for propagation, processing (including wait for I- frame) and transmission times.

35 Examples HDLC High-level Data Link Control V.42 Modem protocol

36 Performance For error free transmission efficiency is close to perfect The window size has to be large enough to be equal to delay-bandwidth product If we lose a frame we have to retransmit all frames (can be costly) For high error rates can be worse than stop and wait

37 Performance We analyze the average performance We assume there are errors We use probabilistic analysis

38 Performance Cost of first transmission Cost of first retransmission t f P f W s t f Cost of all retransmissions P f 2 W s t f t f + i =1 P f i W s t f = t f + P f 1 P f W s t f

39 Performance Expected time for go-back-n t gbn = t f + P f 1 P f W s t f

40 Performance Efficiency of go-back-n η gbn = n f n o t gbn R

41 Selective Repeat ARQ Go-back-n is simple Extremely important a couple of decades ago Good performance if error rates low Performance quickly declines with increasing error rate

42 Selective Repeat ARQ The improvement over stop-n-wait was due to more buffers at the transmitter We can introduce more buffers at the receiver as well Introduce a few minor changes in the protocol

43 Selective Repeat Frames Sent And acked S last S recent S last W s 1 Frames Received And acked R next S last R next S last R next S recent... S last +W s 1...

44 Selective Repeat ARQ A1 A2 N2 A2 A5 A6 A7 A8 A9

45 Maximum Window Sizes Has to be 2 m-1. If it was 2 m -1 the receiver would not know if it is a repeat or new frame.

46 TCP Transmission Control Protocol A variant of Selective Repeat Has 32 bit seq. Number with random start