Principles of Reliable Data Transfer

Transcription

1 Principles of Reliable Data Transfer 1

2 Reliable Delivery Making sure that the packets sent by the sender are correctly and reliably received by the receiver amid network errors, i.e., corrupted/lost packets Can be implemented at LL, NL or TL of the protocol stack. Totally a design choice When and why should this be used? Link Layer Rarely done over twisted-pair or fiber optic links Usually done over lossy links (wireless) for performance improvement (versus correctness) in P2P links Network/Transport Layers Necessary if the application requires the data to be reliably delivered to the receiver, e.g., file transfer 2

3 Reliable Delivery: Service Model RDT_Send Reliable Data Transfer Protocol (Sender Side) Deliver_Data Reliable Data Transfer Protocol (Receiver Side) UDT_Send unreliable channel Reliable, In-order Delivery RDT_Receive Reliable, In-order delivery Typically done when reliability is implemented at the transport layer, e.g., TCP Example application: File transfer 3

4 We ll: Reliable Delivery: Assumptions Consider only unidirectional data transfer A sender sending packets to a receiver Bidirectional communication is a simple extension, where there are 2 sender/receiver pairs Start with simple a protocol and make it complex as we continue 4

5 RDT over Unreliable Channel RDT_Send Reliable Data Transfer Protocol (Sender Side) UDT_Send Unreliable channel Deliver_Data Reliable Data Transfer Protocol (Receiver Side) RDT_Receive Channel may flip bits in packets/lose packets The received packet may have been corrupted during transmission, or dropped at an intermediate router due to buffer overflow The question: how to recover from errors? ACKs, NACKs, Timeouts Next 5

6 RDT over Unreliable Channel Two fundamental mechanisms to accomplish reliable delivery over Unreliable Channels Acknowledgements (ACK), Negative ACK (NACK) Small control packets (header without any data) that a protocol sends back to its peer saying that it has received an earlier packet (positive ACK) or that it has not received a packet (NACK). Sent by the receiver to the sender Timeouts Set by the sender for each transmitted packet If an ACK is received before the timer expires, then the packet has made it to the receiver If the timeout occurs, the sender assumes that the packet is lost (corrupted) and retransmits the packet 6

7 ARQ The general strategy of using ACKs (NACKs) and timeouts to implement reliable delivery is called Automatic Repeat request (ARQ) 3 ARQ Mechanisms for Reliable Delivery Stop and Wait Concurrent Logical Channels Sliding Window 7

8 Timeout Stop and Wait Simplest ARQ protocol Sender: Send a packet Stop and wait until an ACK arrives If received ACK, send the next packet If timeout, ReTransmit the same packet Receiver: When you receive a packet correctly, send an ACK Time Sender Receiver 8

9 Timeout Timeout Timeout Timeout Timeout Timeout Recovering from Error Time Packet lost ACK lost Early timeout Does this protocol work? When an ACK is lost or a early timeout occurs, how does the receiver know whether the packet is a retransmission or a new packet? Use sequence numbers: Both Packets and ACKs 9

10 Timeout Timeout Timeout Timeout Timeout Timeout Stop & Wait with Seq #s Time Packet lost ACK lost Early timeout Sequence # in packet is finite -- how big should it be? One bit won t send Pkt #1 until received ACK for Pkt #0 10

11 Performance of Stop and Wait Can only send one packet per round trip example: 1 Gbps link, 15 ms e-e prop. delay, 1KB packet: first packet bit transmitted, t = 0 last packet bit transmitted, t = L / R sender receive r RTT first packet bit arrives last packet bit arrives, send ACK ACK arrives, send next packet, t = RTT + L / R U sender = L / R.008 = RTT + L / R = = 0.027% microsec microsec 1KB pkt every 30 msec -> 33kB/sec throughput over 1 Gbps link network protocol limits use of physical resources! 11

12 Pipelining: Increasing Utilization Pipelining: sender allows multiple, in-flight, yet-to-be-acknowledged pkts without waiting for first to be ACKed to keep the pipe full Capacity of the Pipe = RTT * BW first packet bit transmitted, t = 0 last bit transmitted, t = L / R sender receiver RTT ACK arrives, send next packet, t = RTT + L / R first packet bit arrives last packet bit arrives, send ACK last bit of 2 nd packet arrives, send ACK last bit of 3 rd packet arrives, send ACK Increase utilization by a factor of 3! U sender = 3 * L / R RTT + L / R = = microsecon 12

13 Sliding Window Protocols Reliable, in-order delivery of packets Sender can send window of up to N, consecutive unack ed packets Receiver makes sure that the packets are delivered in-order to the upper layer 2 Generic Versions Go-Back-N Selective Repeat 13

14 Sliding Window: Generic Sender/Receiver States Sender Receiver Last ACK Received (LAR) Last Packet Sent (LPS) Next Packet Expected (NPE) Last Packet Acceptable (LPA) Sender Window Size Receiver Window Size Sent & Acked Sent Not Acked Received & Acked Acceptable Packet OK to Send Not Usable Not Usable 14

15 Sliding Window- Sender Side The sender maintains 3 variables Sender Window Size (SWS) Upper bound on the number of in-flight packets Last Acknowledgement Received (LAR) Last Packet Sent (LPS) We want LPS LAR <= SWS <= SWS LAR LPS 15

16 Sliding Window- Receiver Side The receiver maintains 3 variables Receiver Window Size (RWS) Upper bound on the number of buffered packets Last Packet Acceptable (LPA) Next Packet Expected (NPE) We want LPS NPE + 1 <= RWS <= RWS NPE LPA 16

17 Go-Back-N Sender Receiver Last ACK Received (LAR) Last Packet Sent (LPS) Next Packet Expected (NPE) Last Packet Acceptable (LPA) SWS = N RWS = 1 packet Sent & Acked Sent Not Acked Received & Acked Acceptable Packet OK to Send Not Usable Not Usable SWS = N: Sender can send up to N consecutive unack ed pkts RWS = 1: Receiver has buffer for just 1 packet Always sends ACK for correctly-rcvd pkt with highest in-order seq # Cumulative ACK Out-of-order pkt: discard & re-ack pkt with highest in-order seq # 17

18 Go-Back-N: Sender Actions Data From Above: Send packets as long as LPS-LAR <= SWS ACK(k): An ACK with seqno = k arrives: If k > LAR then, increase LAR until LAR hits a packet for which ACK has not arrived yet, or LAR == LPS Send packet(s) as long as LPS-LAR <= SWS Associate a timer with the oldest packet sent Single timer for all packets in transit Timeout: Retransmit ALL packets that have been previously sent, but not yet ACKed Therefore the name Go-Back-N 18

19 Go-Back-N: Receiver Actions A packet with seqno arrives: If seqno == NPE then // in-order packet Deliver the packet to the upper layer Send an ACK for pkt# = seqno This is called cumulative ACK scheme ACKs not only the current packet, but also all packets before it If seqno!= NPE then // out of order packet Since sequence # of the last packet received is NPE 1, send an ACK for pkt# = NPE-1 Still using cumulative ACK scheme. 19

20 GBN in action (SWS = 4) 20

21 Why use GBN? Very simple receiver GBN: Last Word Why NOT use GBN? Throwing away out-of-order packets at the receiver results in extra transmissions, thus lowering the channel utilization: The channel may become full of retransmissions of old packets rather than useful new packets Can we do better? Yes: Buffer out-of-order packets at the receiver and do Selective Repeat (Retransmissions) at the sender 21

22 Selective Repeat Sender Receiver Last ACK Received (LAR) Last Packet Sent (LPS) Next Packet Expected (NPE) Last Packet Acceptable (LPA) SWS = N RWS = N Sent & Acked Sent Not Acked Received & Acked Acceptable Packet OK to Send Not Usable Not Usable SWS = RWS = N consecutive packets: Sender can send up to N consecutive unack ed pkts, Receiver can buffer up to N consecutive packets Receiver individually acknowledges all correctly received pkts buffers pkts, as needed, for eventual in-order delivery to upper layer Sender only resends pkts for which ACK not received sender timer for each unacked pkt 22

23 sender data from above : if next available seq # in window, send pkt timeout(k): resend pkt k, restart timer ACK(k) in [LAR+1, LPS] Mark pkt k as received if k == LAR +1 then, advance LAR to next unacked pkt # Selective repeat receiver pkt k in [NPE, LPA] send ACK(k) out-of-order (k!= NPE): buffer in-order (k == NPE): deliver (also deliver buffered, in-order pkts), advance NPE to next notyet-received pkt pkt k in [NPE-N, LPA-1] Send ACK(k) otherwise: ignore 23

24 Selective repeat in action 24

25 Selective Repeat: Sequence Numbers How large do sequence numbers need to be? Must be able to detect wrap-around Depends on sender/receiver window size E.g. Assume SWS = RWS = 7. Also assume that we use 3-bit sequence numbers, i.e., 0..7 Sender sends frames 0..6 Assume receiver received all these frames successfully BUT all ACKs are lost Receiver expects 7,0..5 Sender timeouts and retransmits old 0..6 Receiver receives these but assumes these are new frames!! It turns out that the sending window size can be no more than half as big as the number of available sequence numbers WS <= (MaxSeqNo +1)/2 25

26 Sliding Window: Last Word Go-Back-N and Selective Repeat are NOT the only sliding Window protocol alternatives Other variations exist: Let SWS = RWS = N and use cumulative ACKs Simplifies the sender: Can have a single timer instead of a timer for each packet in transit This is in fact what TCP does! Let SWS = RWS = N, and use Negative ACKs Can be used when the channel is pretty reliable, i.e., packet loss is very rare Only notify the sender when something goes wrong 26