Congestion Control in TCP

Transcription

1 Congestion Control in TCP Antonio Carzaniga Faculty of Informatics University of Lugano November 11, 2014

2 Outline Intro to congestion control Input rate vs. output throughput Congestion window Congestion avoidance Slow start Fast recovery

3 Understanding Congestion A router behaves a lot like a kitchen sink

4 Understanding Congestion A router behaves a lot like a kitchen sink max throughput T

5 Understanding Congestion A router behaves a lot like a kitchen sink λ 1 = T/2 max throughput T throughput= T/2

6 Understanding Congestion A router behaves a lot like a kitchen sink λ 1 = T/2 λ 2 = T/2 max throughput T throughput= T

7 Understanding Congestion A router behaves a lot like a kitchen sink λ 3 = T/2 λ 1 = T/2 λ 2 = T/2 max throughput T throughput= T

8 Understanding Congestion A router behaves a lot like a kitchen sink λ 3 = T/2 λ 1 = T/2 λ 2 = T/2 max throughput T throughput= T

9 Understanding Congestion A router behaves a lot like a kitchen sink λ 3 = T/2 λ 1 = T/2 λ 2 = T/2 max throughput T throughput= T

10 Understanding Congestion A router behaves a lot like a kitchen sink λ 3 = T/2 λ 1 = T/2 λ 2 = T/2 max throughput T throughput= T

11 Understanding Congestion A router behaves a lot like a kitchen sink λ 3 = T/2 λ 1 = T/2 λ 2 = T/2 max throughput T throughput= T

12 Queuing Delay Antonio Carzaniga

13 Queuing Delay Total latency is the sum of link latency, processing time, and the time that a packet spends in the input queue L= TX + CPU + q where q = q /T

14 Queuing Delay Total latency is the sum of link latency, processing time, and the time that a packet spends in the input queue L= TX + CPU + q where q = q /T Ideal case: constant input data rate λ in < T In this case the q = 0, because q =0

15 Queuing Delay Total latency is the sum of link latency, processing time, and the time that a packet spends in the input queue L= TX + CPU + q where q = q /T Ideal case: constant input data rate λ in < T In this case the q = 0, because q =0 Extreme case: constant input data rate λ in > T In this case q =(λ in T)t and therefore q = λ in T T t

16 Queuing Delay Antonio Carzaniga

17 Queuing Delay Steady-state queuing delay 0 λ in < T q = λ in T T t λ in > T

18 Queuing Delay Steady-state queuing delay 0 λ in < T q = λ in T T t λ in > T q λ in T ideal input flow λ in constant

19 Queuing Delay Steady-state queuing delay 0 λ in < T q = λ in T T t λ in > T q q λ in T ideal input flow λ in constant λ in T realistic input flow λ in variable

20 Queuing Delay Antonio Carzaniga

21 Queuing Delay Conclusion: as the input rateλ in approaches the maximum throughput T, packets will experience very long delays

22 Queuing Delay Conclusion: as the input rateλ in approaches the maximum throughput T, packets will experience very long delays More realistic assumptions and models finite queue length (buffers) in routers effects of retransmission overhead

23 Queuing Delay Conclusion: as the input rateλ in approaches the maximum throughput T, packets will experience very long delays More realistic assumptions and models finite queue length (buffers) in routers effects of retransmission overhead T λ out λ in

24 Queuing Delay Conclusion: as the input rateλ in approaches the maximum throughput T, packets will experience very long delays More realistic assumptions and models finite queue length (buffers) in routers effects of retransmission overhead T congestion λ out λ in

25 What to Do? What to do when the network is congested and queues are full? λ 3 = T/2 λ 1 = T/2 λ 2 = T/2 max throughput T throughput= T

26 What to Do? What to do when the network is congested and queues are full? λ 1 = T/2 max throughput T throughput= T

27 What to Do? What to do when the network is congested and queues are full? λ 1 = T/2 max throughput T throughput= T

28 What to Do? What to do when the network is congested and queues are full? λ 1 = T/2 max throughput T throughput= T

29 What to Do? What to do when the network is congested and queues are full? λ 1 = T/2 max throughput T throughput= T

30 Congestion Control (in TCP) Antonio Carzaniga

31 Congestion Control (in TCP) Approach: the sender limits its output rate according to the status of the network the sender output rate becomes (part of) the input rate for the network (λ in )

32 Congestion Control (in TCP) Approach: the sender limits its output rate according to the status of the network the sender output rate becomes (part of) the input rate for the network (λ in ) Issues

33 Congestion Control (in TCP) Approach: the sender limits its output rate according to the status of the network the sender output rate becomes (part of) the input rate for the network (λ in ) Issues how does the sender measure the status of the network? i.e., how does the sender detect congestion? Antonio Carzaniga

34 Congestion Control (in TCP) Approach: the sender limits its output rate according to the status of the network the sender output rate becomes (part of) the input rate for the network (λ in ) Issues how does the sender measure the status of the network? i.e., how does the sender detect congestion? how does the sender effectively limit its output rate? Antonio Carzaniga

35 Congestion Control (in TCP) Approach: the sender limits its output rate according to the status of the network the sender output rate becomes (part of) the input rate for the network (λ in ) Issues how does the sender measure the status of the network? i.e., how does the sender detect congestion? how does the sender effectively limit its output rate? how should the sender modulate its output rate? i.e., what algorithm should the sender use to decrease or increase its output rate? Antonio Carzaniga

36 Detecting Congestion Antonio Carzaniga

37 Detecting Congestion If all traffic is correctly acknowledged, then the sender assumes (quite correctly) that there is no congestion

38 Detecting Congestion If all traffic is correctly acknowledged, then the sender assumes (quite correctly) that there is no congestion Congestion means that the queue of one or more routers between the sender and the receiver overflow the visible effect is that some segments are dropped

39 Detecting Congestion If all traffic is correctly acknowledged, then the sender assumes (quite correctly) that there is no congestion Congestion means that the queue of one or more routers between the sender and the receiver overflow the visible effect is that some segments are dropped Therefore the server assumes that the network is congested when it detects a segment loss time out (i.e., no ACK) multiple acknowledgements (i.e., NACK)

40 Congestion Window The sender maintains a congestion window W

41 Congestion Window The sender maintains a congestion window W The congestion window limits the amount of bytes that the sender pushes into the network before blocking waiting for acknowledgments

42 Congestion Window The sender maintains a congestion window W The congestion window limits the amount of bytes that the sender pushes into the network before blocking waiting for acknowledgments where LastByteSent LastByteAcked W W = min(congestionwindow, ReceiverWindow)

43 Congestion Window The sender maintains a congestion window W The congestion window limits the amount of bytes that the sender pushes into the network before blocking waiting for acknowledgments where LastByteSent LastByteAcked W W = min(congestionwindow, ReceiverWindow) The resulting maximum output rate is roughly λ= W 2L

44 Congestion Control How does TCP modulate its output rate?

45 Congestion Control How does TCP modulate its output rate? Additive-increase and multiplicative-decrease

46 Congestion Control How does TCP modulate its output rate? Additive-increase and multiplicative-decrease Slow start

47 Congestion Control How does TCP modulate its output rate? Additive-increase and multiplicative-decrease Slow start Reaction to timeout events

48 Additive-Increase/Multiplicative-Decrease Antonio Carzaniga

49 Additive-Increase/Multiplicative-Decrease How W is reduced: at every loss event, TCP halves the congestion window

50 Additive-Increase/Multiplicative-Decrease How W is reduced: at every loss event, TCP halves the congestion window e.g., suppose the window size W is currently 20Kb, and a loss is detected TCP reduces W to 10Kb

51 Additive-Increase/Multiplicative-Decrease How W is reduced: at every loss event, TCP halves the congestion window e.g., suppose the window size W is currently 20Kb, and a loss is detected TCP reduces W to 10Kb How W is increased: at every (good) acknowledgment, TCP increments W by 1MSS/W, so as to increase W by MSS every round-trip time 2L. This process is called congestion avoidance

52 Additive-Increase/Multiplicative-Decrease How W is reduced: at every loss event, TCP halves the congestion window e.g., suppose the window size W is currently 20Kb, and a loss is detected TCP reduces W to 10Kb How W is increased: at every (good) acknowledgment, TCP increments W by 1MSS/W, so as to increase W by MSS every round-trip time 2L. This process is called congestion avoidance e.g., suppose W = and MSS= 1460, then the sender increases W to after 10 acknowledgments acknowledgments

53 Additive-Increase/Multiplicative-Decrease Window size W over time W Time

54 Slow Start What is the initial value of W?

55 Slow Start What is the initial value of W? The initial value of W is MSS, that is 1 segment, which is quite low for modern networks

56 Slow Start What is the initial value of W? The initial value of W is MSS, that is 1 segment, which is quite low for modern networks To get quickly to a good throughput level, TCP increases its sending rate exponentially for its first growth phase, up to a slow-start threshold (ssthresh)

57 Slow Start What is the initial value of W? The initial value of W is MSS, that is 1 segment, which is quite low for modern networks To get quickly to a good throughput level, TCP increases its sending rate exponentially for its first growth phase, up to a slow-start threshold (ssthresh) After the threshold, TCP proceeds with its linear push

58 Slow Start What is the initial value of W? The initial value of W is MSS, that is 1 segment, which is quite low for modern networks To get quickly to a good throughput level, TCP increases its sending rate exponentially for its first growth phase, up to a slow-start threshold (ssthresh) After the threshold, TCP proceeds with its linear push This process is called slow start because of the small initial value of W

59 Timeouts vs. NACKs As we know, three duplicate ACKs are interpreted as a NACK

60 Timeouts vs. NACKs As we know, three duplicate ACKs are interpreted as a NACK Both timeouts and NACKs signal a loss, but they say different things about the status of the network

61 Timeouts vs. NACKs As we know, three duplicate ACKs are interpreted as a NACK Both timeouts and NACKs signal a loss, but they say different things about the status of the network A timeout indicates congestion

62 Timeouts vs. NACKs As we know, three duplicate ACKs are interpreted as a NACK Both timeouts and NACKs signal a loss, but they say different things about the status of the network A timeout indicates congestion Three (duplicate) ACKs suggest that the network is still able to deliver segments along that path

63 Timeouts vs. NACKs As we know, three duplicate ACKs are interpreted as a NACK Both timeouts and NACKs signal a loss, but they say different things about the status of the network A timeout indicates congestion Three (duplicate) ACKs suggest that the network is still able to deliver segments along that path So, TCP reacts differently to a timeout and to a triple duplicate ACKs

64 Assuming the current window size is W = W Timeouts vs. NACKs

65 Assuming the current window size is W = W Timeout go back to W= MSS set ssthresh= W/2 run slow start up to W = ssthresh then proceed with congestion avoidance Timeouts vs. NACKs

66 Assuming the current window size is W = W Timeout go back to W= MSS set ssthresh= W/2 run slow start up to W = ssthresh then proceed with congestion avoidance Timeouts vs. NACKs NACK (i.e., triple duplicate-ack) set ssthresh= W/2 cut W in half: W = W/2 run congestion avoidance, ramping up W linearly This is called fast recovery

67 Sender Behavior W Time

68 Sender Behavior W MSS Time

69 Sender Behavior W NACK MSS Time

70 Sender Behavior W NACK MSS Time

71 Sender Behavior timeout W NACK MSS Time

72 Sender Behavior timeout W NACK MSS Time

73 Sender Behavior timeout W NACK MSS Time

74 Sender Behavior timeout W NACK MSS Time

75 Sender Behavior NACK timeout W NACK MSS Time

76 Sender Behavior NACK W NACK timeout NACK MSS Time

77 Sender Behavior NACK W NACK timeout NACK MSS Time

78 Sender Behavior NACK W NACK timeout NACK MSS Time

79 Sender Behavior SS=slow start CA=congestion avoidance SS CA SS CA CA CA NACK W NACK timeout NACK MSS Time