TCP Deconstructed

Slides: TCP Deconstructed

[Start Here (html)] [Slide Sorter (html)]

Download: [Postscript:] [PDF: CalTech200210.pdf]

Page 1: TCP Deconstructed

Page 2: Outline

Page 3: Background: What is the Internet?

Page 4: Background: What is a protocol?

Page 5: Example: Fat TCP

Page 6: "Solved" problems

Page 7: Solved problem #1: Robust self clock

Page 8: Solved problem #2: Autotuning

Page 9: The next step: the Web100 Project

Page 10: The Wizard Gap

Page 11: The Wizard Gap

Page 12: Why? TCP "tuning" requires experts

Page 13: TCP tuning is really debugging

Page 14: TCP tuning is painful debugging

Page 15: Web100 Instrumentation (TCP MIB)

Page 16: General Comments on Instrumentation

Page 17: New Autotuning

Page 18: General comments on flow control (Rwin)

Page 19: Path diagnosis

Page 20: Implications of the TCP Model

Page 21: Change the CC model?

Page 22: What about MTU?

Page 23: Some MTU observations

Page 24: MTU history revisited

Page 25: Fairness and MTU

Page 26: Hard constraints

Page 27: TCP's Success Disaster

Page 28: TCP is Byte Oriented

Page 29: 32 bit Sequence Limit

Page 30: Conclusion

Generated by MagicPoint