Pushing up Performance for Everyone

Slides: Pushing up
for Everyone

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


Page 1: Pushing up

Page 2: Pushing up performance

Page 3: The Wizard Gap

Page 4: The Wizard Gap

Page 5: TCP "tuning" requires experts

Page 6: TCP tuning is painful debugging

Page 7: Web100

Page 8: When there is a problem, just ask TCP

Page 9: TCP triage instrument

Page 10: Diagnosing path bottlenecks

Page 11: Diagnosing path bugs

Page 12: Localizing path problems

Page 13: Sender and receiver bottlenecks

Page 14: Web100 Summary

Page 15: Current Web100 status

Page 16: end of part 1

Page 17: The Other Problem:

Page 18: The TCP model

Page 19: Fitting the model 1

Page 20: Fitting the model 2

Page 21: Fitting the model 3

Page 22: Fitting the model 4

Page 23: Fitting the model 5

Page 24: The other problem is the loss rate

Page 25: end of part 2

Page 26: Implications to users

Page 27: How well are users doing today?

Page 28: We need independent measures of

Page 29: The General Plan

Page 30: Passive measurements

Page 31: Window size measurements

Page 32: Observed window sizes

Page 33: Observed window sizes

Page 34: Receiver window sizes

Page 35: Receiver window sizes

Page 36: end of part 3

Page 37: Implications to the network

Page 38: Danger - new load levels

Page 39: Why not before Web100?

Page 40: We expect to see line rate traffic

Page 41: Problems

Page 42: No deployed queue management

Page 43: No deployed QoS

Page 44: Pervasive broken link layers

Page 45: No models for traffic sharing

Page 46: Shared gigabit problem

Page 47: No SLA quantification

Page 48: New load levels

Page 49: The common theme

Page 50: A Traffic Dynamics Testbed

Page 51: end of part 4

Generated by MagicPoint