Transport Layer (TCP/UDP)

Transcription

1 Transport Layer (TCP/UDP)

2 Recall the protocol stack Application Transport Network Link Physical Programs that use network service Provides end-to-end data delivery Send packets over multiple networks Send frames over one or more links Send bits using signals Computer Networks 2

3 Transport layer Provides end-to-end connectivity to applications Client Transport Network Server Transport Network Host Host CSE 461 University of Washington 3

4 Transport layer protocols Provide different kinds of data delivery across the network to applications Messages Bytestream Unreliable Datagrams (UDP) Reliable Streams (TCP) CSE 461 University of Washington 4

5 Comparison of Internet transports TCP is full-featured, UDP is a glorified packet TCP (Streams) Connections Bytes are delivered once, reliably, and in order Arbitrary length content Flow control matches sender to receiver Congestion control matches sender to network UDP (Datagrams) Datagrams Messages may be lost, reordered, duplicated Limited message size Can send regardless of receiver state Can send regardless of network state CSE 461 University of Washington 5

6 Socket API Simple abstraction to use the network The network API (really Transport service) used to write all Internet apps Part of all major OSes and languages; originally Berkeley (Unix) ~1983 Supports both Internet transport services (Streams and Datagrams) CSE 461 University of Washington 6

7 Socket API (2) Sockets let apps attach to the local network at different ports Socket, Port #1 Socket, Port #2 CSE 461 University of Washington 7

8 Socket API (3) Same API used for Streams and Datagrams Only needed for Streams To/From for Datagrams Primitive Meaning SOCKET Create a new communication endpoint BIND Associate a local address (port) with a socket LISTEN Announce willingness to accept connections ACCEPT Passively establish an incoming connection CONNECT Actively attempt to establish a connection SEND(TO) Send some data over the socket RECEIVE(FROM) Receive some data over the socket CLOSE Release the socket CSE 461 University of Washington 8

9 Ports Application process is identified by the tuple IP address, transport protocol, and port Ports are 16-bit integers representing local mailboxes that a process leases Servers often bind to well-known ports <1024, require administrative privileges Clients often assigned ephemeral ports Chosen by OS, used temporarily CSE 461 University of Washington 9

10 Some Well-Known Ports Port Protocol Use 20, 21 FTP File transfer 22 SSH Remote login, replacement for Telnet 25 SMTP 80 HTTP World Wide Web 110 POP-3 Remote access 143 IMAP Remote access 443 HTTPS Secure Web (HTTP over SSL/TLS) 543 RTSP Media player control 631 IPP Printer sharing CSE 461 University of Washington 10

11 Topics Service models Socket API and ports Datagrams, Streams User Datagram Protocol (UDP) Connections (TCP) Sliding Window (TCP) Flow control (TCP) Retransmission timers (TCP) Congestion control (TCP) CSE 461 University of Washington 11

12 UDP

13 User Datagram Protocol (UDP) Used by apps that don t want reliability or bytestreams Like what? CSE 461 University of Washington 13

14 User Datagram Protocol (UDP) Used by apps that don t want reliability or bytestreams Voice-over-IP DNS DHCP (If application wants reliability and messages then it has work to do!) CSE 461 University of Washington 14

15 Datagram Sockets Client (host 1) Time Server (host 2) request reply CSE 461 University of Washington 15

16 Datagram Sockets (2) Client (host 1) Time Server (host 2) 1: socket 4: sendto request 1: socket 2: bind 3: recvfrom* 5: recvfrom* reply 6: sendto 7: close 7: close *= call blocks CSE 461 University of Washington 16

17 UDP Buffering Application App App App Ports Transport (UDP) Message queues Network (IP) Port Mux/Demux packet CSE 461 University of Washington 17

18 UDP Header Uses ports to identify sending and receiving application processes Datagram length up to 64K Checksum (16 bits) for reliability CSE 461 University of Washington 18

19 UDP Header (2) Optional checksum covers UDP segment and IP pseudoheader Checks key IP fields (addresses) Value of zero means no checksum CSE 461 University of Washington 19

20 TCP

21 TCP TCP Consists of 3 primary phases: Connection Establishment (Setup) Sliding Windows/Flow Control Connection Release (Teardown)

22 Connection Establishment Both sender and receiver must be ready before we start the transfer of data Need to agree on a set of parameters e.g., the Maximum Segment Size (MSS) This is signaling It sets up state at the endpoints Like dialing for a telephone call CSE 461 University of Washington 22

23 Three-Way Handshake Used in TCP; opens connection for data in both directions Each side probes the other with a fresh Initial Sequence Number (ISN) Sends on a SYNchronize segment Echo on an ACKnowledge segment Chosen to be robust even against delayed duplicates Active party (client) Passive party (server) CSE 461 University of Washington 23

24 Three-Way Handshake (2) Three steps: Client sends SYN(x) Server replies with SYN(y)ACK(x+1) Client replies with ACK(y+1) SYNs are retransmitted if lost Sequence and ack numbers carried on further segments Active party (client) 1 SYN (SEQ=x) 2 SYN (SEQ=y, ACK=x+1) Time 3 (SEQ=x+1, ACK=y+1) Passive party (server) CSE 461 University of Washington 24

25 Three-Way Handshake (3) Suppose delayed, duplicate copies of the SYN and ACK arrive at the server! Improbable, but anyhow Active party (client) SYN (SEQ=x) Passive party (server) (SEQ=x+1, ACK=z+1) CSE 461 University of Washington 25

26 Three-Way Handshake (4) Suppose delayed, duplicate copies of the SYN and ACK arrive at the server! Improbable, but anyhow Connection will be cleanly rejected on both sides J Active party (client) X REJECT SYN (SEQ=x) SYN (SEQ=y, ACK=x+1) (SEQ=x+1, ACK=z+1) Passive party (server) X REJECT CSE 461 University of Washington 26

27 TCP Connection State Machine Captures the states ([]) and transitions (->) A/B means event A triggers the transition, with action B Both parties run instances of this state machine

28 TCP Connections (2) Follow the path of the client:

29 TCP Connections (3) And the path of the server:

30 TCP Connections (4) Again, with states Active party (client) CLOSED SYN_SENT 1 SYN (SEQ=x) Passive party (server) CLOSED LISTEN ESTABLISHED 2 SYN (SEQ=y, ACK=x+1) 3 (SEQ=x+1, ACK=y+1) SYN_RCVD Time ESTABLISHED CSE 461 University of Washington 30

31 TCP Connections (5) Finite state machines are a useful tool to specify and check the handling of all cases that may occur TCP allows for simultaneous open i.e., both sides open instead of the client-server pattern Try at home to confirm it works J CSE 461 University of Washington 31

32 Connection Release Orderly release by both parties when done Delivers all pending data and hangs up Cleans up state in sender and receiver Key problem is to provide reliability while releasing TCP uses a symmetric close in which both sides shutdown independently CSE 461 University of Washington 32

33 TCP Connection Release Two steps: Active sends FIN(x), passive ACKs Passive sends FIN(y), active ACKs FINs are retransmitted if lost Active party Passive party Each FIN/ACK closes one direction of data transfer CSE 461 University of Washington 33

34 TCP Connection Release (2) Two steps: Active sends FIN(x), passive ACKs Passive sends FIN(y), active ACKs FINs are retransmitted if lost Each FIN/ACK closes one direction of data transfer Active party FIN (SEQ=x) 1 (SEQ=y, ACK=x+1) FIN (SEQ=y, ACK=x+1) 2 (SEQ=x+1, ACK=y+1) Passive party CSE 461 University of Washington 34

35 TCP Connection State Machine Both parties run instances of this state machine CSE 461 University of Washington 35

36 TCP Release Follow the active party CSE 461 University of Washington 36

37 TCP Release (2) Follow the passive party CSE 461 University of Washington 37

38 TCP Release (3) Again, with states Active party ESTABLISHED FIN_WAIT_1 FIN_WAIT_2 FIN (SEQ=x) TIME_WAIT (timeout) CLOSED CLOSED 1 (SEQ=y, ACK=x+1) FIN (SEQ=y, ACK=x+1) 2 (SEQ=x+1, ACK=y+1) Passive party ESTABLISHED CLOSE_WAIT LAST_ACK CSE 461 University of Washington 38

39 TIME_WAIT State Wait a long time after sending all segments and before completing the close Two times the maximum segment lifetime of 60 seconds Why? CSE 461 University of Washington 39

40 TIME_WAIT State Wait a long time after sending all segments and before completing the close Two times the maximum segment lifetime of 60 seconds Why? ACK might have been lost, in which case FIN will be resent for an orderly close Could otherwise interfere with a subsequent connection CSE 461 University of Washington 40

41 Flow Control

42 Flow control goal Match transmission speed to reception capacity Otherwise data will be lost

43 ARQ: Automatic repeat query ARQ with one message at a time is Stop-and-Wait Timeout Sender Receiver Frame 0 ACK 0 Time Frame 1 ACK 1 CSE 461 University of Washington 43

44 Limitation of Stop-and-Wait It allows only a single message to be outstanding from the sender: Fine for LAN (only one frame fits in network anyhow) Not efficient for network paths with longer delays CSE 461 University of Washington 44

45 Limitation of Stop-and-Wait (2) Example: B=1 Mbps, D = 50 ms, 10kb packets RTT (Round Trip Time) = 2D = 100 ms How many packets/sec? 10 Usage efficiency if packet is 10kb? (10,000 x 10) / (1 x 10 6 ) = 10% What is the efficient B=10 Mbps? 1% CSE 461 University of Washington 45

46 Sliding Window Generalization of stop-and-wait Allows W packets to be outstanding Can send W packets per RTT (=2D) Pipelining improves performance Need W=2BD to fill network path CSE 461 University of Washington 46

47 Sliding Window (2) What W will use the network capacity with 10kb packets? Ex: B=1 Mbps, D = 50 ms 2BD = 2 x 10 6 x 50/1000 = 100 Kb W = 100 kb/10 = 10 packets Ex: What if B=10 Mbps? W = 100 packets CSE 461 University of Washington 47

48 Sliding Window Protocol Many variations, depending on how buffers, acknowledgements, and retransmissions are handled Go-Back-N Simplest version, can be inefficient Selective Repeat More complex, better performance CSE 461 University of Washington 48

49 Sender Sliding Window Sender buffers up to W segments until they are acknowledged LFS=LAST FRAME SENT, LAR=LAST ACK REC D Sends while LFS LAR W Sliding Window W=5 Available.. 5Acked Unacked Unavailable LAR LFS seq. number CSE 461 University of Washington 49

50 Sender Sliding Window (2) Transport accepts another segment of data from the Application... Transport sends it (LFS LAR à 5) Sliding Window W=5 Sent.. 5Acked Unacked Unavailable LAR LFS seq. number CSE 461 University of Washington 50

51 Sender Sliding Window (3) Next higher ACK arrives from peer Window advances, buffer is freed LFS LAR à 4 (can send one more) Sliding Window W=5 Available.. 5Acked Unacked Unavailable LAR LFS seq. number CSE 461 University of Washington 51

52 Receiver Sliding Window Go-Back-N Receiver keeps only a single packet buffer for the next segment State variable, LAS = LAST ACK SENT On receive: If seq. number is LAS+1, accept and pass it to app, update LAS, send ACK Otherwise discard (as out of order) CSE 461 University of Washington 52

53 Receiver Sliding Window Selective Repeat Receiver passes data to app in order, and buffers out-oforder segments to reduce retransmissions ACK conveys highest in-order segment, plus hints about outof-order segments Ex: I got everything up to 42 (LAS), and got 44, 45 TCP uses a selective repeat design; we ll see the details later CSE 461 University of Washington 53

54 Receiver Sliding Window Selective Repeat (2) Buffers W segments, keeps state variable LAS = LAST ACK SENT On receive: Buffer segments [LAS+1, LAS+W] Send app in-order segments from LAS+1, and update LAS Send ACK for LAS regardless CSE 461 University of Washington 54

55 Sender Sliding Window Selective Repeat Keep normal sliding window If out-of-order ACK arrives Send LAR again! Sliding Window W=5 Ack Arrives Out of Order!.. 5Acked Unacked Unavailable LAR again LFS seq. number CSE 461 University of Washington 55

56 Sender Sliding Window Selective Repeat (2) Keep normal sliding window If in-order ACK arrives Move window and LAR, send more messages Sliding Window W=5 In-order ack arrives Now Available.. 5Acked Unacked LAR LFS seq. number CSE 461 University of Washington 56

57 Sliding Window Retransmissions Go-Back-N uses a single timer to detect losses On timeout, resends buffered packets starting at LAR+1 Selective Repeat uses a timer per unacked segment to detect losses On timeout for segment, resend it Hope to resend fewer segments CSE 461 University of Washington 57

58 Sequence Numbers Need more than 0/1 for Stop-and-Wait but how many? For Selective Repeat: 2W seq numbers W for packets, plus W for earlier acks For Go-Back-N: W+1 sequence numbers Typically implement seq. number with an N-bit counter that wraps around at 2 N 1 E.g., N=8:, 253, 254, 255, 0, 1, 2, 3, CSE 461 University of Washington 58

59 Sequence Time Plot Seq. Number Transmissions (at Sender) Delay (=RTT/2) Acks (at Receiver) Time CSE 461 University of Washington 59

60 Sequence Time Plot (2) Go-Back-N scenario Seq. Number Time CSE 461 University of Washington 60

61 Sequence Time Plot (3) Retransmissions Seq. Number Loss Timeout Time CSE 461 University of Washington 61

62 ACK Clocking

63 Sliding Window ACK Clock Typically, the sender does not know B or D Each in-order ACK advances the sliding window and lets a new segment enter the network ACKs clock data segments Data Ack CSE 461 University of Washington 63

64 Benefit of ACK Clocking Consider what happens when sender injects a burst of segments into the network Queue Fast link Slow (bottleneck) link Fast link CSE 461 University of Washington 64

65 Benefit of ACK Clocking (2) Segments are buffered and spread out on slow link Segments spread out Fast link Slow (bottleneck) link Fast link CSE 461 University of Washington 65

66 Benefit of ACK Clocking (3) ACKs maintain the spread back to the original sender Slow link Acks maintain spread CSE 461 University of Washington 66

67 Benefit of ACK Clocking (4) Sender clocks new segments with the spread Now sending at the bottleneck link without queuing! Segments spread Queue no longer builds Slow link CSE 461 University of Washington 67

68 Benefit of ACK Clocking (4) Helps run with low levels of loss and delay! The network smooths out the burst of data segments ACK clock transfers this smooth timing back to sender Subsequent data segments are not sent in bursts so do not queue up in the network CSE 461 University of Washington 68

69 TCP Uses ACK Clocking TCP uses a sliding window because of the value of ACK clocking Sliding window controls how many segments are inside the network TCP only sends small bursts of segments to let the network keep the traffic smooth CSE 461 University of Washington 69

70 Problem Sliding window has pipelining to keep network busy What if the receiver is overloaded? Arg Big Iron Streaming video Wee Mobile CSE 461 University of Washington 70

71 Receiver Sliding Window Consider receiver with W buffers LAS=LAST ACK SENT app pulls in-order data from buffer with recv() call Sliding Window W=5.. 5Finished Acceptable Too 3 high LAS seq. number CSE 461 University of Washington 71

72 Receiver Sliding Window (2) Suppose the next two segments arrive but app does not call recv() W=5.. 5Finished Acceptable Too 3 high LAS seq. number CSE 461 University of Washington 72

73 Receiver Sliding Window (3) Suppose the next two segments arrive but app does not call recv() LAS rises, but we can t slide window! W=5.. 5Finished 6 7 Acked Too 3 high LAS seq. number CSE 461 University of Washington 73

74 Receiver Sliding Window (4) Further segments arrive (in order) we fill buffer Must drop segments until app recvs! W=5 Nothing Acceptable!.. 5Finished 6 7 Acked Too 3 high LAS seq. number CSE 461 University of Washington 74

75 Receiver Sliding Window (5) App recv() takes two segments Window slides (phew) W=5 Acceptable.. 5Finished Acked LAS seq. number CSE 461 University of Washington 75

76 Flow Control Avoid loss at receiver by telling sender the available buffer space WIN=#Acceptable, not W (from LAS) W=5 Acceptable.. 5Finished Acked LAS seq. number CSE 461 University of Washington 76

77 Flow Control (2) Sender uses lower of the sliding window and flow control window (WIN) as the effective window size W=3 Acceptable.. 5Finished 6 7 Acked Too 3 high LAS seq. number CSE 461 University of Washington 77

78 Flow Control (3) TCP-style example SEQ/ACK sliding window Flow control with WIN SEQ + length < ACK+WIN 4KB buffer at receiver Circular buffer of bytes CSE 461 University of Washington 78

79 Topic How to set the timeout for sending a retransmission Adapting to the network path Lost? Network CSE 461 University of Washington 79

80 Retransmissions With sliding window, detecting loss with timeout Set timer when a segment is sent Cancel timer when ack is received If timer fires, retransmit data as lost Retransmit! CSE 461 University of Washington 80

81 Timeout Problem Timeout should be just right Too long à inefficient network capacity use Too short à spurious resends waste network capacity But what is just right? Easy to set on a LAN (Link) Short, fixed, predictable RTT Hard on the Internet (Transport) Wide range, variable RTT CSE 461 University of Washington 81

82 Example of RTTs BCNàSEAàBCN Round Trip Time (ms) Seconds CSE 461 University of Washington 82

83 Example of RTTs (2) Round Trip Time (ms) BCNàSEAàBCN Variation due to queuing at routers, changes in network paths, etc. Propagation (+transmission) delay 2D Seconds CSE 461 University of Washington 83

84 Example of RTTs (3) Round Trip Time (ms) Timer too high! Need to adapt to the network conditions Timer too low! Seconds CSE 461 University of Washington 84

85 Adaptive Timeout Smoothed estimates of the RTT (1) and variance in RTT (2) Update estimates with a moving average 1. SRTT N+1 = 0.9*SRTT N + 0.1*RTT N+1 2. Svar N+1 = 0.9*Svar N + 0.1* RTT N+1 SRTT N+1 Set timeout to a multiple of estimates To estimate the upper RTT in practice TCP Timeout N = SRTT N + 4*Svar N CSE 461 University of Washington 85

86 Example of Adaptive Timeout RTT (ms) SRTT Svar Seconds CSE 461 University of Washington 86

87 Example of Adaptive Timeout (2) Early timeout Timeout (SRTT + 4*Svar) RTT (ms) Seconds CSE 461 University of Washington 87

88 Adaptive Timeout (2) Simple to compute, does a good job of tracking actual RTT Little headroom to lower Yet very few early timeouts Turns out to be important for good performance and robustness CSE 461 University of Washington 88

89 Congestion

90 TCP to date: We can set up and tear connections Connection establishment and release handshakes Keep the sending and receiving buffers from overflowing (flow control) What s missing?

91 Network Congestion A traffic jam in the network Later we will learn how to control it What s the hold up? Network CSE 461 University of Washington 91

92 Congestion Collapse in the 1980s Early TCP used fixed size window (e.g., 8 packets) Initially fine for reliability But something happened as the network grew Links stayed busy but transfer rates fell by orders of magnitude! CSE 461 University of Washington 92

93 Nature of Congestion Routers/switches have internal buffering Input Output Input Buffer Fabric Output Buffer CSE 461 University of Washington 93

94 Nature of Congestion (2) Simplified view of per port output queues Typically FIFO (First In First Out), discard when full Router Router = (FIFO) Queue Queued Packets CSE 461 University of Washington 94

95 Nature of Congestion (3) Queues help by absorbing bursts when input > output rate But if input > output rate persistently, queue will overflow This is congestion Congestion is a function of the traffic patterns can occur even if every link has the same capacity CSE 461 University of Washington 95

96 Effects of Congestion What happens to performance as we increase load?

97 Effects of Congestion (2) What happens to performance as we increase load?

98 Effects of Congestion (3) As offered load rises, congestion occurs as queues begin to fill: Delay and loss rise sharply with load Throughput < load (due to loss) Goodput << throughput (due to spurious retransmissions) None of the above is good! Want network performance just before congestion CSE 461 University of Washington 98

99 TCP Tahoe/Reno TCP extensions and features we will study: AIMD Fair Queuing Slow-start Fast Retransmission Fast Recovery CSE 461 University of Washington 100

100 TCP Timeline 3-way handshake (Tomlinson, 75) Origins of TCP (Cerf & Kahn, 74) TCP/IP flag day (BSD Unix 4.2, 83) TCP and IP (RFC 791/793, 81) Congestion collapse Observed, TCP Reno (Jacobson, 90) TCP Tahoe (Jacobson, 88) Pre-history Congestion control CSE 461 University of Washington 101

101 TCP Timeline (2) Delay based TCP Reno (Jacobson, 90) ECN (Floyd, 94) TCP Vegas (Brakmo, 93) Router support TCP with SACK (Floyd, 96) TCP New Reno (Hoe, 95) FAST TCP (Low et al., 04) Background TCP BIC (Linux, 04 TCP LEDBAT (IETF 08) Compound TCP (Windows, 07) TCP CUBIC (Linux, 06) Classic congestion control Diversification CSE 461 University of Washington 102

102 Bandwidth Allocation Important task for network is to allocate its capacity to senders Good allocation is both efficient and fair Efficient: most capacity is used but there is no congestion Fair: every sender gets a reasonable share of the network CSE 461 University of Washington 103

103 Efficiency vs. Fairness Cannot always have both! Example network with traffic: AàB, BàC and AàC How much traffic can we carry? A B C 1 1 CSE 461 University of Washington 104

104 Efficiency vs. Fairness (2) If we care about fairness: Give equal bandwidth to each flow AàB: ½ unit, BàC: ½, and AàC, ½ Total traffic carried is 1 ½ units A B C 1 1 CSE 461 University of Washington 105

105 Efficiency vs. Fairness (3) If we care about efficiency: Maximize total traffic in network AàB: 1 unit, BàC: 1, and AàC, 0 Total traffic rises to 2 units! A B C 1 1 CSE 461 University of Washington 106

106 Fairness What s a fair bandwidth allocation? The max-min fair allocation CSE 461 University of Washington 107

107 The Slippery Notion of Fairness Why is equal per flow fair anyway? AàC uses more network resources than AàB or BàC Host A sends two flows, B sends one Not productive to seek exact fairness More important to avoid starvation A node that cannot use any bandwidth Equal per flow is good enough CSE 461 University of Washington 108

108 Generalizing Equal per Flow Bottleneck for a flow of traffic is the link that limits its bandwidth Where congestion occurs for the flow For AàC, link A B is the bottleneck A B C 1 10 Bottleneck CSE 461 University of Washington 109

109 Generalizing Equal per Flow (2) Flows may have different bottlenecks For AàC, link A B is the bottleneck For BàC, link B C is the bottleneck Can no longer divide links equally A B C 1 10 CSE 461 University of Washington 110

110 Max-Min Fairness Intuitively, flows bottlenecked on a link get an equal share of that link Max-min fair allocation is one that: Increasing the rate of one flow will decrease the rate of a smaller flow This maximizes the minimum flow CSE 461 University of Washington 111

111 Max-Min Fairness (2) To find it given a network, imagine pouring water into the network 1. Start with all flows at rate 0 2. Increase the flows until there is a new bottleneck in the network 3. Hold fixed the rate of the flows that are bottlenecked 4. Go to step 2 for any remaining flows CSE 461 University of Washington 112

112 Max-Min Example Example: network with 4 flows, link bandwidth = 1 What is the max-min fair allocation? CSE 461 University of Washington 113

113 Max-Min Example (2) When rate=1/3, flows B, C, and D bottleneck R4 R5 Fix B, C, and D, continue to increase A Bottleneck Bottleneck CSE 461 University of Washington 114

114 Max-Min Example (3) When rate=2/3, flow A bottlenecks R2 R3. Done. Bottleneck Bottleneck CSE 461 University of Washington 115

115 Max-Min Example (4) End with A=2/3, B, C, D=1/3, and R2 R3, R4 R5 full Other links have extra capacity that can t be used, linksxample: network with 4 flows, links equal bandwidth What is the max-min fair allocation? CSE 461 University of Washington 116

116 Adapting over Time Allocation changes as flows start and stop Time CSE 461 University of Washington 117

117 Adapting over Time (2) Flow 1 slows when Flow 2 starts Flow 1 speeds up when Flow 2 stops Flow 3 limit is elsewhere Time CSE 461 University of Washington 118

118 Recall the goals of bandwidth allocation Want to allocate capacity to senders Network layer provides feedback Transport layer adjusts offered load A good allocation is efficient and fair CSE 461 University of Washington 119

119 Why is Bandwidth Allocation hard? Number of senders and their offered load changes Senders may be limited in other ways Other parts of network or by applications Network is distributed; no single party has an overall picture of its state CSE 461 University of Washington 120

120 Bandwidth Allocation Models Open loop versus closed loop Open: reserve bandwidth before use Closed: use feedback to adjust rates Host versus Network support Who is sets/enforces allocations? Window versus Rate based How is allocation expressed? TCP is a closed loop, host-driven, and window-based CSE 461 University of Washington 123

121 Bandwidth Allocation Models (2) We ll study closed-loop, host-driven, and windowbased too Network layer returns feedback on current allocation to senders At least tells if there is congestion Transport layer adjusts sender s behavior via window in response How senders adapt is a control law CSE 461 University of Washington 124

122 Additive Increase Multiplicative Decrease AIMD is a control law hosts can use to reach a good allocation Hosts additively increase rate while network not congested Hosts multiplicatively decrease rate when congested Used by TCP Let s explore the AIMD game CSE 461 University of Washington 125

123 AIMD Game Hosts 1 and 2 share a bottleneck But do not talk to each other directly Router provides binary feedback Tells hosts if network is congested Host 1 Host Router Bottleneck 1 Rest of Network CSE 461 University of Washington 126

124 AIMD Game (2) Each point is a possible allocation Host 1 1 Congested Fair Optimal Allocation Efficient 0 1 Host 2 CSE 461 University of Washington 127

125 AIMD Game (3) AI and MD move the allocation Host 1 1 Congested Additive Increase Multiplicative Decrease Fair, y=x Optimal Allocation Efficient, x+y=1 0 1 Host 2 CSE 461 University of Washington 128

126 AIMD Game (4) Play the game! Host 1 1 Congested A starting point Fair Efficient 0 1 Host 2 CSE 461 University of Washington 129

127 AIMD Game (5) Always converge to good allocation! Host 1 1 Congested A starting point Fair Efficient 0 1 Host 2 CSE 461 University of Washington 130

128 AIMD Sawtooth Produces a sawtooth pattern over time for rate of each host This is the TCP sawtooth (later) Host 1 or 2 s Rate Multiplicative Decrease Additive Increase Time CSE 461 University of Washington 131

129 AIMD Properties Converges to an allocation that is efficient and fair when hosts run it Holds for more general topologies Other increase/decrease control laws do not! (Try MIAD, MIMD, MIAD) Requires only binary feedback from the network CSE 461 University of Washington 132

130 Feedback Signals Several possible signals, with different pros/cons We ll look at classic TCP that uses packet loss as a signal Signal Example Protocol Pros / Cons Packet loss Packet delay Router indication TCP NewReno Cubic TCP (Linux) Compound TCP (Windows) TCPs with Explicit Congestion Notification Hard to get wrong Hear about congestion late Other events can cause loss Hear about congestion early Need to infer congestion Hear about congestion early Require router support CSE 461 University of Washington 133

131 Slow Start (TCP Additive Increase)

132 What we know about TCP so far Connection establishment 3-way handshake Connection release FIN/ACK for each direction Flow control Sliding windows Congestion control basics Max-min fairness, AIMD This lecture: TCP congestion control in practice

133 TCP congestion control overview Follows AIMD control law for a good allocation Sender uses a congestion window or cwnd to set its rate ( cwnd/rtt) Sender uses loss as network congestion signal Designed to work across a very large range of rates and RTTs CSE 461 University of Washington 136

134 TCP Slow Start Problem We want to quickly near the right rate, cwnd IDEAL, but it varies greatly Fixed sliding window doesn t adapt and is rough on the network (loss!) Additive Increase with small bursts adapts cwnd gently, but might take a long time to become efficient CSE 461 University of Washington 137

135 Slow-Start Solution Start by doubling cwnd every RTT Exponential growth (1, 2, 4, 8, 16, ) Start slow, quickly reach large values Window (cwnd) Fixed Slow-start AI Time 138

136 Slow-Start Solution (2) Eventually packet loss will occur when the network is congested Loss timeout tells us cwnd is too large Next time, switch to AI beforehand Slowly adapt cwnd near right value In terms of cwnd: Expect loss for cwnd C 2BD+queue Use ssthresh = cwnd C /2 to switch to AI CSE 461 University of Washington 139

137 Slow-Start Solution (3) Combined behavior, after first time Most time spent near right value Window cwnd C cwnd IDEAL ssthresh Fixed Slow-start AI phase AI Time 140

138 Slow-Start (Doubling) Timeline Increment cwnd by 1 packet for each ACK CSE 461 University of Washington 141

139 Additive Increase Timeline Increment cwnd by 1 packet every cwnd ACKs (or 1 RTT) CSE 461 University of Washington 142

140 TCP Tahoe (Implementation) Initial slow-start (doubling) phase Start with cwnd = 1 (or small value) cwnd += 1 packet per ACK Later Additive Increase phase cwnd += 1/cwnd packets per ACK Roughly adds 1 packet per RTT Switching threshold (initially infinity) Switch to AI when cwnd > ssthresh Set ssthresh = cwnd/2 after loss Begin with slow-start after timeout CSE 461 University of Washington 143

141 Timeout Misfortunes Why do a slow-start after timeout? Instead of MD cwnd (for AIMD) Timeouts are sufficiently long that the ACK clock will have run down Slow-start ramps up the ACK clock We need to detect loss before a timeout to get to full AIMD CSE 461 University of Washington 144

142 Fast Recovery (TCP Multiplicative Decrease)

143 Inferring Loss from ACKs TCP uses a cumulative ACK Carries highest in-order seq. number Normally a steady advance Duplicate ACKs give us hints about what data hasn t arrived Tell us some new data did arrive, but it was not next segment Thus the next segment may be lost CSE 461 University of Washington 146

144 Fast Retransmit Treat three duplicate ACKs as a loss Retransmit next expected segment Some repetition allows for reordering, but still detects loss quickly Ack CSE 461 University of Washington 147

145 Fast Retransmit (2) Third duplicate ACK, so send 14 ACK jumps after loss is repaired Ack 10 Ack 11 Ack 12 Ack 13 Ack 13 Ack 13 Ack 13 Ack Ack Data 20 Data 14 Data 14 was lost earlier, but got 15 to 20 Retransmission fills in the hole at 14 CSE 461 University of Washington 148

146 Fast Retransmit (3) It can repair single segment loss quickly, typically before a timeout However, we have quiet time at the sender/receiver while waiting for the ACK to jump And we still need to MD cwnd CSE 461 University of Washington 149

147 Inferring Non-Loss from ACKs Duplicate ACKs also give us hints about what data has arrived Each new duplicate ACK means that some new segment has arrived It will be the segments after the loss Thus advancing the sliding window will not increase the number of segments stored in the network CSE 461 University of Washington 150

148 Fast Recovery First fast retransmit, and MD cwnd Then pretend further duplicate ACKs are the expected ACKs Lets new segments be sent for ACKs Reconcile views when the ACK jumps Ack CSE 461 University of Washington 151

149 Fast Recovery (2) Third duplicate ACK, so send 14 Set ssthresh, cwnd = cwnd/2 More ACKs advance window; may send segments before jump Ack 12 Ack 13 Ack 13 Ack 13 Ack 13 Ack 13 Ack 13 Ack Exit Fast Recovery Data 20 Data 14 Data 21 Data 22 Data 14 was lost earlier, but got 15 to 20 Retransmission fills in the hole at 14 CSE 461 University of Washington 152

150 Fast Recovery (3) With fast retransmit, it repairs a single segment loss quickly and keeps the ACK clock running This allows us to realize AIMD No timeouts or slow-start after loss, just continue with a smaller cwnd TCP Reno combines slow-start, fast retransmit and fast recovery Multiplicative Decrease is ½ CSE 461 University of Washington 153

151 TCP Reno TCP sawtooth MD of ½, no slow-start ACK clock running CSE 461 University of Washington 154

152 TCP Reno, NewReno, and SACK Reno can repair one loss per RTT Multiple losses cause a timeout NewReno further refines ACK heuristics Repairs multiple losses without timeout Selective ACK (SACK) is a better idea Receiver sends ACK ranges so sender can retransmit without guesswork CSE 461 University of Washington 155

153 Network-Assisted Congestion Control

154 Congestion Avoidance vs. Control Classic TCP drives the network into congestion and then recovers Needs to see loss to slow down Would be better to use the network but avoid congestion altogether! Reduces loss and delay But how can we do this? CSE 461 University of Washington 157

155 Feedback Signals Delay and router signals can let us avoid congestion Signal Example Protocol Pros / Cons Packet loss Packet delay Router indication Classic TCP Cubic TCP (Linux) Compound TCP (Windows) TCPs with Explicit Congestion Notification Hard to get wrong Hear about congestion late Other events can cause loss Hear about congestion early Need to infer congestion Hear about congestion early Require router support CSE 461 University of Washington 158

156 ECN (Explicit Congestion Notification) Router detects the onset of congestion via its queue When congested, it marks affected packets (IP header) CSE 461 University of Washington 159

157 ECN (2) Marked packets arrive at receiver TCP receiver informs TCP sender of the congestion CSE 461 University of Washington 160

158 ECN (3) Advantages: Routers deliver clear signal to hosts Congestion is detected early, no loss No extra packets need to be sent Disadvantages: Routers and hosts must be upgraded CSE 461 University of Washington 161

159 Recent transport layer developments (1) QUIC

160 Recent transport layer developments (2) Multipath TCP (MPTCP) By Aclarembeau - Own work, CC BY-SA 4.0,