Bandwidth Estimation for TCP Sources and its Application

1. Bandwidth Estimation for TCP Sources and its Application Prepared for QoS IP 2003 R. G. Garroppo, S.Giordano, M. Pagano, G. Procissi, R. Secchi Department of Information Engineering University of Pisa

2. Introduction • TCP is the most used reliable and connection oriented transport service • TCP is called to operate in different network environment • TCP Westwood proposal enhances TCP performance and fairness, expecially in mixed wired/wireless environment

3. Outline • Bandwith Estimation • Algorithm • Numerical validation • Accuracy in estimation • Application to TCP • Performance Evaluation • Fairness and Friendliness • Conclusion

4. Bandwidth Estimation Technique • BWE is computed at the sender side by processing ACK time series • Estimator’s design assumes a simple network model (single bottleneck approximation) • Available Bandwidth is computed by comparing packet inter-departure time of packet and from inter-arrival time of ACKs

5. Bandwidth Estimator Scheme If queueing occurs at the bottleneck L packet size µ bottleneck capacity λcross traffic rate

6. Linear-Regression technique • Collect a sample of RTT and ACK inter-arrival time, only if: • Perform a linear over sample pairs: • Computes bandwidth estimation, by performing linear-regression:

7. Bandwidth Estimation Numerical Validation UDP Source 10Mbs, 10ms Loss free channel NS-2 Simulation Topology Lossy link 2Mbs, 40ms TCP Source Lossy channel

8. Bandwidth Estimation Numerical Validation (a) Lossfree channel (b) Lossy channel Disturb Traffic flowing in the opposite direction

9. Bandwidth Estimation Bandwidth Estimation Accuracy BWE Accuracy (with concurrent CBR/UDP Traffic)

10. Application to TCP Algorithm flow chart 3 DupACKs Received Timeout expires ssthresh :=pipesize cwnd := 1 ssthresh := pipesize cwnd > ssthresh no ssthresh < 2 no yes yes cwnd := ssthresh ssthresh := 2

11. Topology of Simulation Application to TCP NS-2 topology Modified TCP Source

12. GoodPut Comparison Application to TCP (a) UDP rate: 100Kbps (b) UDP rate: 500Kbps TCP Througput vs Loss Probability on lossy link

13. Application to TCP Completion Time Comparison Average Completion Time ( 60 packets long flows )

14. Fairness and Friendliness concepts • Fairness: identical connection sharing a common resource achieve the same performance • Friedliness:A friendly protocol doesn’t hurt standard protocol more than a homogeneous protocol

15. Modified NewReno-Fairness Application to TCP (a) Lossfree channel (b) Lossy channel Goodput achieved by a couple of connections sharing a common link. Different CBR/UDP disturb traffic rate are considered.

16. Application to TCP Modified NewReno Friendliness Modified NewReno Friendliness towards NewReno (b) Lossy channel (a) Lossfree channel

17. Application to TCP Modified NewReno Friendliness NewReno vs. Mod. NewReno in lossy channel

18. Conclusion • We propose an end-to-end bandwidth estimation technique for TCP connection. • Extensive simulation validate the effectiveness of the algorithm under various network conditions. • The algorithm has been applied to TCP congestion control mechanism. • The new protocol performance is assessed through simulations, showing clear improvement and compatibility towards NewReno.