1. Khiem Lam Jimmy Vuong Andrew Yang
TCP Over Wireless in NS2
Khiem Lam
Jimmy Vuong
Andrew Yang

2. Introduction Background
TCP flavors: Reno, NewReno, SACK
Wireless: WLAN, WWAN, Satellite
NS2 – Network simulator
Problem: TCP performance degrades over wireless networks
Motivation: As wireless becomes more and more prevalent, TCP, being the primary transport protocol over the internet, starts to bottleneck over these links.

3. Milestones
Set up Linux, install NS2, and perform the Marc Greis tutorial
Study the performance of TCP Reno, NewReno, and SACK on wired networks.
Study the performances in wireless networks
Develop (in theory) algorithms to improve throughput
Implement new algorithms and obtain results

4. Milestone 1: Preliminary Work
Setup NS2
Problems
Linux
Required libraries
Manually install package
Learn TCL and NS2 functionality
Marc Greis Tutorial
Wireless very different
Only UDP
Learn TCP

5. Milestone 2: Wired Simulation
100s Simulation Time
Constant bit rate source
Reno
New Reno
SACK
Throughput
9.966Mbps
9.975Mbps
9.098Mbps

6. Milestone 3: Wireless
Typical Values for Bandwidth and Delay from Various Networks
Bandwidth
Delay
WLAN a
21.8Mbps-32.76Mbps
1ms-5ms
WLAN b
4.48Mbps-6.73Mbps
WWAN EDGE
108kbps-162kbps
200ms-300ms
WWAN Evdo
0.8Mbps-1.2Mbps
160ms-240ms
Satellite
1.2Mbps-1.8Mbps
480ms-720ms

7. Wireless Simulation Setup
100s Simulation time
Average of 5 runs per data point
Bandwidth vary between 0.05Mbps and 10Mbps
Delay vary between 5ms and 1000ms
Analysis Setup
Calculate total throughput from total bytes transferred at simulation end
Load and plot values in MATLAB
Simulations took ~30-45minutes per TCP flavor to get good graph resolution

8. Wireless (5ms Delay)

9. Wireless (1s Delay)

10. Wireless (1.05mbps Bandwidth)

11. Wireless (Sequence Numbers)
10% loss
10Mb
150ms

12. Analysis and Proposal Loss and delay strongly affect the throughput.
2 part attack
Primary focus on Satellite networks
Large delay and large error
Small window sizes stay small because delays and lost ACKs cause timeouts.
Delay
Window Plus (WP): increase window faster with greater delays
Larger increase in window for every ACK if large delay

13. Proposal (contd.) Loss Key: differentiate between loss and congestion.
Error Monitoring (EM):
If there is loss & small deviation in RTT don’t slow down
Shout by Chance (SbC)
Look at unacknowledged packets
If small deviation in RTT  every unACK packet has chance of being retransmitted.
If large deviation in RTT  congestion

14. Implementation Major changes to code Check for error loss
Added delay logger
Check for losses

15. TCP-SACK with WP (low delay)

16. TCP-SACK with WP (high delay)

17. TCP-SACK with WP (Delay)

18. TCP-SACK with EM (low delay)

19. TCP-SACK with EM (high delay)

20. TCP-SACK with EM (Delay)

21. TCP-SACK w/ EM+WP (Low Delay)

22. TCP-SACK w/ EM+WP (High Delay)

23. TCP-SACK with EM+WP (Delay)

24. TCP-SACK w/ EM+WP (Seq.)
10% loss
10Mb
150ms

25. Summary and Future Work
TCP-Sack with EM+WP performs much better than current TCP implementations
Future work
Further work into fairness
Simulation with 2+ flows
More investigation into timeouts
Apply more improvements
Implement SbC
Modify NewReno instead of SACK

26. Appendix Complete Collection of Graphs
Throughput vs. (Bandwidth, Delay, Error) in Various Conditions
Sequence Number vs. Time

27. Appendix A
Wireless Simulation of TCP SACK, Reno, and NewReno

28. Wireless (5ms Delay)

29. Wireless (150ms Delay)

30. Wireless (400ms Delay)

31. Wireless (1s Delay)

32. Wireless (50kbps Bandwidth)

33. Wireless (1.05mbps Bandwidth)

34. Wireless (6.52mbps Bandwidth)

35. Wireless (10mbps Bandwidth)

36. Wireless (Sequence Numbers)
10% loss
10Mb
150ms

37. Wireless (Low Delay)

38. Wireless (Low Delay)

39. Wireless (Low Delay)

40. Wireless (Low Delay)

41. Wireless (High Delay)

42. Wireless (High Delay)

43. Wireless (High Delay)

44. Wireless (High Delay)

45. Appendix B
Wireless Simulation of TCP SACK, Reno, and TCP+WP

46. TCP-SACK+WP (5ms Delay)

47. TCP-SACK+WP (150ms Delay)

48. TCP-SACK+WP (400ms Delay)

49. TCP-SACK+WP (1s Delay)

50. SACK+WP (0.05mbps Bandwidth)

51. SACK+WP (1.05mbps Bandwidth)

52. SACK+WP (6.52mbps Bandwidth)

53. SACK+WP (10mbps Bandwidth)

54. Appendix C
Wireless Simulation of TCP SACK, Reno, and TCP+EM

55. TCP-SACK+EM (5ms Delay)

56. TCP-SACK+EM (150ms Delay)

57. TCP-SACK+EM (400ms Delay)

58. TCP-SACK+EM (1s Delay)

59. SACK+EM (0.05mbps Bandwidth)

60. SACK+EM (1.05mbps Bandwidth)

61. SACK+EM (6.52mbps Bandwidth)

62. SACK+EM (10mbps Bandwidth)

63. Appendix D
Wireless Simulation of TCP SACK, Reno, and TCP+EM+WP

64. TCP-SACK+EM+WP (5ms Delay)

65. TCP-SACK+EM+WP (150ms Delay)

66. TCP-SACK+EM+WP (400ms Delay)

67. TCP-SACK+EM+WP (1s Delay)

68. SACK+EM+WP (0.05mbps BW)

69. SACK+EM+WP (1.05mbps BW)

70. SACK+EM+WP (6.52mbps BW)

71. SACK+EM+WP (10mbps BW)

72. SACK+EM+WP (Sequence)
10% loss
10Mb
150ms

73. SACK+EM+WP (Low Delay)

74. SACK+EM+WP (Low Delay)

75. SACK+EM+WP (Low Delay)

76. SACK+EM+WP (Low Delay)

77. SACK+EM+WP (High Delay)

78. SACK+EM+WP (High Delay)

79. SACK+EM+WP (High Delay)

80. SACK+EM+WP (High Delay)