The document discusses various TCP congestion control algorithms including TCP Reno, NewReno, and SACK. It provides details on how each algorithm performs congestion control including congestion window adjustments and fast retransmit/recovery. It also discusses the deployment of these algorithms and introduces TCP-friendly rate control (TFRC) as an equation-based congestion control for unreliable transports like UDP.

1. TCP"
Reno, NewReno, SACK
4 September 2009
CS5229, Semester 1, 2009/10
1

2. TCP Reno
4 September 2009
CS5229, Semester 1, 2009/10
2

3. 14
cwnd
ssthresh
12
10
8
6
4
2
0
4 September 2009
CS5229, Semester 1, 2009/10
3

4. new ack:"
if (cwnd < sstresh)"
cwnd += 1"
else"
cwnd += 1/cwnd
4 September 2009
CS5229, Semester 1, 2009/10
5

5. timeout:"
retransmit 1st unacked"
ssthresh = cwnd/2"
cwnd = 1"
4 September 2009
CS5229, Semester 1, 2009/10
6

6. 3rd duplicate ACK:"
fast retransmission"
(ie, retransmit 1st unack) "
fast recovery"
(details today)"
ssthresh = cwnd = cwnd/2
4 September 2009
CS5229, Semester 1, 2009/10
7

7. TCP’s rule"
send more packets if "
L + cwnd > H"
[ L .. H-1 ] are outstanding packets
4 September 2009
CS5229, Semester 1, 2009/10
10

8. 4 September 2009
CS5229, Semester 1, 2009/10
11

9. Data Packets
ACK Packets
(received up to this seq number)
Congestion Window
Sequence
Number
Time
4 September 2009
CS5229, Semester 1, 2009/10
12

10. 3rd dup ack:"
retransmit 1st unacked"
ssthresh = cwnd/2"
cwnd = cwnd/2
4 September 2009
CS5229, Semester 1, 2009/10
13

11. Incorrectly clamping the congestion window
too soon (without fast recovery)
X
4 September 2009
CS5229, Semester 1, 2009/10
14

12. fast recovery:"
keep the pipe occupied"
4 September 2009
CS5229, Semester 1, 2009/10
15

13. X
4 September 2009
CS5229, Semester 1, 2009/10
16

14. 4 September 2009
CS5229, Semester 1, 2009/10
17

15. X
X
4 September 2009
CS5229, Semester 1, 2009/10
18

16. TCP Reno timeout with
multiple losses in a window
4 September 2009
CS5229, Semester 1, 2009/10
19

17. TCP NewReno
4 September 2009
CS5229, Semester 1, 2009/10
20

18. X
X
4 September 2009
CS5229, Semester 1, 2009/10
21

19. 4 September 2009
CS5229, Semester 1, 2009/10
22

20. 3rd dup ack:"
retransmit 1st unacked"
ssthresh = cwnd/2"
cwnd = cwnd/2 + 3"
remember highest
4 September 2009
CS5229, Semester 1, 2009/10
25

21. “complete” ack:"
(all are acked) "
cwnd = ssthresh
4 September 2009
CS5229, Semester 1, 2009/10
26

22. “partial” ack:"
(acknowledge n packets)"
retransmit"
cwnd = cwnd - n + 1
4 September 2009
CS5229, Semester 1, 2009/10
27

23. Note: RFC2581/RFC2582 give the
accurate/gory details. Simplified version
is presented here (eg. cwnd vs FlightSize,
update of cwnd upon partial ACK).
4 September 2009
CS5229, Semester 1, 2009/10
29

24. What does a dup
ACK tell us?
4 September 2009
CS5229, Semester 1, 2009/10
30

25. “Coarse Feedback”
4 September 2009
CS5229, Semester 1, 2009/10
31

26. TCP SACK
4 September 2009
CS5229, Semester 1, 2009/10
32

27. Use TCP header
options to report
received segments.
4 September 2009
CS5229, Semester 1, 2009/10
33

28. SACK Blocks:"
1st block - report most recently
received segments"
subsequent blocks - repeat most
recent previous blocks
4 September 2009
CS5229, Semester 1, 2009/10
34

29. pipe: num of outstanding
packets in the pipe."
send only if pipe < cwnd
4 September 2009
CS5229, Semester 1, 2009/10
35

30. scoreboard: which
packets have been
received?
4 September 2009
CS5229, Semester 1, 2009/10
36

31. X
X
4 September 2009
CS5229, Semester 1, 2009/10
37

32. 3rd dup ack:"
pipe = cwnd - 3"
retransmit 1st unacked"
ssthresh = cwnd/2"
cwnd = cwnd/2 + 3
4 September 2009
CS5229, Semester 1, 2009/10
38

33. subsequent dup ack:"
cwnd++"
pipe--"
if pipe < cwnd"
send packet, pipe++
4 September 2009
CS5229, Semester 1, 2009/10
39

34. “partial” ack:"
retransmit"
cwnd = cwnd - n + 1"
pipe -= 2"
if pipe < cwnd"
send packet, pipe++
4 September 2009
CS5229, Semester 1, 2009/10
40

35. Idea of SACK:"
Which packet has left the network?"
Where is the gap?"
Decouple when to send and what to
send.
4 September 2009
CS5229, Semester 1, 2009/10
43

36. TCP SACK recovers faster
than NewReno with
multiple losses in a window.
4 September 2009
CS5229, Semester 1, 2009/10
44

37. Deployment
4 September 2009
CS5229, Semester 1, 2009/10
57

38. Feb 2004
NewReno
Reno
8%
Tahoe
15%
77%
4 September 2009
CS5229, Semester 1, 2009/10
58

39. 70% "
SACK capable
4 September 2009
CS5229, Semester 1, 2009/10
59

40. TCP BIC/CUBIC"
Linux 2.6.x
4 September 2009
CS5229, Semester 1, 2009/10
63

41. Compound TCP "
MS Windows Vista
4 September 2009
CS5229, Semester 1, 2009/10
64

42. TFRC"
equation-based"
congestion control
4 September 2009
CS5229, Semester 1, 2009/10
65

43. Not everyone uses TCP
4 September 2009
CS5229, Semester 1, 2009/10
66

44. UDP:"
Media streaming"
Gaming"
VoIP
4 September 2009
CS5229, Semester 1, 2009/10
67

45. Why not congestion controlled?"
1. UDP has low delay, no need
full reliability. UDP "
is not congestion controlled.
4 September 2009
CS5229, Semester 1, 2009/10
68

46. Why not congestion controlled?"
2. No incentive. OTOH, there
are incentives NOT to use
congestion control.
4 September 2009
CS5229, Semester 1, 2009/10
69

47. Sally Floyd, h+p://www.icir.org/floyd/tcp_unfriendly.html
4 September 2009
CS5229, Semester 1, 2009/10
70

48. “Unresponsive Flows”
4 September 2009
CS5229, Semester 1, 2009/10
71

49. Bad: lead to unfairness
and congestion collapse.
4 September 2009
CS5229, Semester 1, 2009/10
72

50. Unfairness: unresponsive
flows consume more bandwidth
than congestion controlled
flows.
4 September 2009
CS5229, Semester 1, 2009/10
73

51. Congestion Collapse:
wasting bandwidth by sending
packets that will be dropped
4 September 2009
CS5229, Semester 1, 2009/10
77

52. Today: "
a TCP-friendly"
unreliable protocol
4 September 2009
CS5229, Semester 1, 2009/10
78

53. Idea: "
send at a rate that a TCP
flow would send
4 September 2009
CS5229, Semester 1, 2009/10
79

54. we can do the AIMD-thing"
at the source, or
4 September 2009
CS5229, Semester 1, 2009/10
80

55. 4 September 2009
CS5229, Semester 1, 2009/10
81

56. equation-based "
congestion control
4 September 2009
CS5229, Semester 1, 2009/10
82

57. steady"
fair"
responsive
4 September 2009
CS5229, Semester 1, 2009/10
83

58. how to determine"
RTT"
tRTO"
p
4 September 2009
CS5229, Semester 1, 2009/10
84

59. p is not packet loss rate"
but "
loss event rate
4 September 2009
CS5229, Semester 1, 2009/10
85

60. 4 September 2009
CS5229, Semester 1, 2009/10
86

61. 4 September 2009
CS5229, Semester 1, 2009/10
87

62. 4 September 2009
CS5229, Semester 1, 2009/10
88

63. RTT can fluctuates
4 September 2009
CS5229, Semester 1, 2009/10
89

64. 4 September 2009
CS5229, Semester 1, 2009/10
90

65. 4 September 2009
CS5229, Semester 1, 2009/10
91

66. 4 September 2009
CS5229, Semester 1, 2009/10
92

67. 4 September 2009
CS5229, Semester 1, 2009/10
93

68. how to initialize?
4 September 2009
CS5229, Semester 1, 2009/10
94

69. slow start"
(until loss occur)"
Tnow = min(2Tprev, 2Trecv)
4 September 2009
CS5229, Semester 1, 2009/10
95

70. no loss history, how?"
solve p given T, RTT
4 September 2009
CS5229, Semester 1, 2009/10
96

71. 4 September 2009
CS5229, Semester 1, 2009/10
97

72. TFRC is now part of
DCCP
4 September 2009
CS5229, Semester 1, 2009/10
98