1 / 30

Adaptive Control for TCP Flow Control

Adaptive Control for TCP Flow Control. Thesis Presentation Amir Maor. Presentation Structure. Introduction AdaVegas intuition AdaVegas simulation results Adaptive mechanism for TCP new Reno Conclusion. Flow Control Existing Solutions TCP NewReno.

selia
Download Presentation

Adaptive Control for TCP Flow Control

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Adaptive Control for TCP Flow Control Thesis Presentation Amir Maor

  2. Presentation Structure • Introduction • AdaVegas intuition • AdaVegas simulation results • Adaptive mechanism for TCP new Reno • Conclusion

  3. Flow Control Existing Solutions TCP NewReno • The sending rate increases continuously until a packet is dropped. • Two Phases • Slow Start – increase exponentially • Congestion Avoidance – increase linearly

  4. NewReno Dynamics

  5. Congestion Avoidance Slow Start  <  Increase Rate Linearly  <  Double Rate << -  >  Exit  >  Decrease Rate Linearly Flow Control Existing Solutions TCP Vegas • Stop increasing rate before swamping the network • Delay  (propagation delay) + (queuing) •  Estimates number of queued packets

  6. Vegas Dynamics

  7. Reno And Vegas Dynamics

  8. Related Work – Chiu & JainAIMD converges to stable and fair operating point

  9. Related Work – Bansal & Barakrishnank+l>0; k>=0 or l>=0

  10. Related Work – Bansal & Barakrishnan

  11. Related Work – Bansal & Barakrishnan

  12. Progress • Introduction • AdaVegas intuition • AdaVegas simulation results • Adaptive mechanism for TCP new Reno • Conclusion

  13. How Adaptive Is Vegas? • Vegas changes the sending rate • Vegas does not change the way it changes the sending rate +1 seg/RTT ; -1 seg/RTT ; -0.5*(rate) • Is this way optimal?

  14. Linear Increase Constant Optimal Value or Painful Compromise ?

  15. Making Vegas Adaptive • The larger the available bandwidth the larger the increase constant • How do we know how large the available bandwidth is? • We don’t ! BUT we can take a pretty good guess by using recent history

  16. Making Vegas AdaptiveAlpha & Beta parameters

  17. AdaVegas Dynamics

  18. Progress • Introduction • AdaVegas intuition • AdaVegas simulation results • Adaptive mechanism for TCP new Reno • Conclusion

  19. SOURCE 1 SOURCE 1 SINK 1 SINK 1 Evaluation criteria 1 msec 100Mb/sec 1 msec 100Mb/sec 1 msec 100Mb/sec 1 msec 100Mb/sec SOURCE 2 SOURCE 2 SINK 2 SINK 2 ON mean time SOURCE 3 SOURCE 3 ROUTER 1 ROUTER 1 300 msec 20Mb/sec 300 msec 20Mb/sec ROUTER 2 ROUTER 2 SINK 3 SINK 3 # users SOURCE N SOURCE N SINK N SINK N Simulation Model • ON/OFF users using heavy tailed distribution • Evaluation criteria: • Line utilization, queue size,loss rate,fairness

  20. Evaluation criteria Results - Utilization ON mean time # users

  21. Evaluation criteria Results – Queue Size ON mean time # users

  22. Evaluation criteria Results – Fairness Index ON mean time # users

  23. Evaluation criteria Results – Loss Rate ON mean time # users

  24. Heterogeneous EnvironmentsDifferent RTT

  25. Progress • Introduction • AdaVegas intuition • AdaVegas simulation results • Adaptive mechanism for TCP new Reno • Conclusion

  26. Import AdaVegas to NewReno? • Relation between increase constant and available bandwidth does not hold in NewReno

  27. NewReno and High Increase Rate

  28. Congestion Avoidance -Barking up the wrong tree?

  29. Progress • Introduction • AdaVegas intuition • AdaVegas simulation results • Adaptive mechanism for TCP new Reno • Conclusion

  30. Conclusion & Future Work • AdaVegas is able to adapt better to changing environments • Research on adaptive mechanisms for NewReno should focus on “Slow Start” as well • Develop adaptive mechanism for NewReno • Make AdaVegas’ increase parameter unbound

More Related