1. 1 TCP-LP: A Distributed Algorithm for Low Priority Data Transfer Aleksandar Kuzmanovic, Edward W. Knightly Department of Electrical and Computer Engineering Rice University IEEE INFOCOM 2003 Presented by Ryan

2. 2 Introduction Service prioritization among different traffic classes E.g. better than best-effort (real-time service) Not easy to deploy in the current Internet TCP-LP (Low Priority) An end-point protocol achieving two-class service prioritization without any support from the network

3. 3 Introduction Objective of TCP-LP Utilizing available bandwidth in a TCP transparent fashion Fair sharing the excess bandwidth among multiple TCP-LP flows (TCP-like fair share) Application of TCP-LP Background file transfer Probing available bandwidth

4. 4 Reference Model Two class hierarchical scheduling model High-priority VS Low-priority class Strict priority service

5. 5 TCP-LP Protocol An end-point congestion control algorithm Early Congestion Indication Congestion Avoidance Policy

6. 6 TCP-LP Early Congestion Indication One-way packet delays as early indicators Smoothed one-way delay (weighted moving average) Early congestion indication condition d – measured one-way delay, γ- delay smoothing parameter, δ- delay threshold

7. 7 TCP-LP Congestion Avoidance Policy Receipt of first early congestion indication  halving the congestion window  entering an inference phase During the inference phase Without increasing the congestion window If receiving another indication  setting the congestion window to 1

8. 8 TCP-LP Congestion Avoidance Policy After the expiration of the inference phase  increasing the congestion window by 1 per RTT (like TCP) Early Congestion Induction

9. 9 Parameter Settings Delay Smoothing, γ= 1/8 (typical value for computing the smoothed RTT for TCP) Delay Threshold, δ= 0.15 Inference Phase Time-out, itt = 3*RTT

10. 10 Simulation Run on NS2 (each run lasts 1000s) Topology Bottleneck link – 1.5Mb/s or 10Mb/s with delay 20ms Other access links – 100Mb/s with delay 2ms

11. 11 Simulation Results FTP and Reverse Background Traffic First Row (excess capacity not available) 2 simultaneous FTP downloads Second Row (excess capacity available) 2 simultaneous FTP downloads 10 TCP flows in the reverse direction

12. 12 Simulation Results Square-wave Background Traffic 1 TCP/TCP-LP flow

13. 13 Simulation Results 10 TCP/TCP-LP flows

14. 14 Simulation Results HTTP Background Traffic Web traffic between Node 0 and 1 FTP connection in the same direction

15. 15 Simulation Multiple Bottlenecks Topology 1 Links 0-1, 1-2 and 2-3 with capacity of 1.5Mb/s Others with capacity of 100Mb/s

16. 16 Simulation Results

17. 17 Simulation Multiple Bottleneck Topology 2 Links capacity – same as Topology 1

18. 18 Simulation Results

19. 19 Conclusion TCP-LP achieves low-priority service without the support of the network Simulations results support its functions Experiments on the Internet should be performed to validate its performance