TCP Westwood Control Model Development and Stability Analysis

1. CS215 TCP Westwood Control Model Developmentand Stability Analysis Hu, Kunzhong khu@seas.ucla.edu Dong, Haibo haibo@seas.ucla.edu Mentor: Wang, Ren Professor: Gerla, Mario March 21, 2001

2. CS215 Outline • Introduction • TCP Tahoe • TCP Reno • TCP Westwood • TCP Bandwidth Estimation • TCP Westwood Throughput Calculation Model • TCPW model performance • Stability and Fairness Analysis • Conclusion and Future Work

3. CS215 TCP Tahoe • Slow Start • Congestion Avoidance

4. Segments wpi wthi 1 Time CS215 TCP Reno • Slow start • Congestion Avoidance • Fast Retransmit • Fast Recovery

5. CS215 Shortcomings of current TCP congestion control • After a sporadic loss, the connection needs several RTTs to be restored to full capacity • It is not possible to distinguish between packet loss caused by congestion (for which a window reduction is in order) and a packet loss caused by wireless interference • The window size selected after a loss may NOT reflect the actual bandwidth available to the connection at the bottleneck

6. CS215 TCP Westwood • Estimation of available bandwidth (BWE): • performed by the source • computed from the arrival rate of ACKs, smoothed through exponential averaging • Use BWE to set the congestion window and the Slow Start threshold

7. CS215 TCP Westwood (Cont.) • When three duplicate ACKs are detected: • set ssthresh=BWE*rtt (instead of ssthresh=cwin/2 as in Reno) • if (cwin > ssthresh) set cwin=ssthresh • When a TIMEOUT expires: • set ssthresh=BWE*rtt (instead of ssthresh=cwnd/2 as in Reno) and cwin=1

8. Throughput Link capacity b2 bk-1 bk b1 bk+1 tk-1 tk+1 tk time RTT Time CS215 CS215 Bandwidth Estimation Expenential Increasement

9. Segments wpi wthi ith cycle 1 Time delayi k*RTT t CS215 TCPW Throughput Calculation Model

10. CS215 Model Performance Comparison of model predication with NS2 simulation for TCP Reno and TCP Westwood  = 45 Mbps,  = 70 ms, MSS = 1340 Byte

11. CS215 Model Performance (cont.) Comparison of window size evolution for TCP Reno and TCP Westwood  = 45 Mbps,  = 70 ms, MSS = 1340 Byte, error rate = 0.01%

12. 1 SES1 SW1 SW2 BW B SES2 DES2 DES1 2 CS215 CS215 TCPW Stability and Fairness Analysis • Two Sources (SES1,SES2), Two Switches (SW1, SW2), Two Destinations (DES1, DES2) • Bandwidth of the shared link is  [seg/sec] • Propagation delay between SESi and DESi is I • Segment size is MSS bytes • Buffer size of SW1 is B

13. w2(t) ws ( w1(t), w2(t) ) Fairness Line we Efficiency Area X Segment Loss Line Efficiency Line w1(t) we ws CS215 CS215 TCPW Stability and Fairness Analysis (Cont.) • Fairness Line, w1(t) = w2(t) • Efficiency Line, link is fully utilized • Segment Loss Line, below, no loss. • we = w1(t) + w2(t) • ws = w1(t) + w2(t) + B TCP Connection Window Size Graph

14. 1 SES1 SW1 SW2 BW B SES2 DES2 DES1 2 CS215 TCPW Stability and Fairness Analysis (Cont.) Throughput of connection j Window size of connection j when ith segment is lost Threshold window size for connection j in ith cycle. Time between i-1th and ith packet loss

15. CS215 Summary and Future Plan • An effective throughput calculation model is developed. • For any given packet loss rate, the evolution of TCP Westwood and TCP Reno window size and connection throughput are properly predicted using this model. • Fairness and stability analysis shows that TCP Westwood is fair but lacks stability • Further attention will be paid on the different methods of calculating the estimated bandwidth and on the throughput calculation model improvement.