文档介绍：“Sizing router buffers”, Appenzeller et al, Sigcomm 2004
“Safe and effective fine-grained TCP…”, Vasudevan et al, Sigcomm 2009
Nick McKeown
CS244 Spring 2012
Lecture 4
Buffer Sizing
2
The Story
10,000
20
# packets
at 10Gb/s
1,000,000
Assume: Large number of desynchronized flows; 100% utilization
Assume: Large number of desynchronized flows; <100% utilization
Intuition
& Proof
Sawtooth
Peak-to-trough
Smoothing of
many sawtooths
Non-bursty
arrivals
Evidence
Simulation,
Single TCP Flow
Simulations,
Measurements
Simulations, Measurements
3
Time evolution of a single TCP flow through a router. Buffer is < 2T*C
Time Evolution of a Single TCP Flow
Time evolution of a single TCP flow through a router. Buffer is 2T*C
4
The Story
10,000
20
# packets
at 10Gb/s
1,000,000
Evidence
Intuition
& Proof
Assume: Large number of desynchronized flows; 100% utilization
Assume: Large number of desynchronized flows; <100% utilization
Sawtooth
Peak-to-trough
Smoothing of
many sawtooths
Non-bursty
arrivals
Simulation,
Single TCP Flow
Simulations,
Experiments
Simulations, Experiments
5
Synchronized Flows
Aggregate window has same dynamics
Therefore buffer occupancy has same dynamics
Rule-of-thumb still holds.
t
6
Probability
Distribution
B
0
Buffer Size
Many TCP Flows
What you said
"I found it quite shocking that before 2004, nobody had bothered putting effort into researching something that was clearly causing huge engineering (and thus financial) hurdles in scaling up the internet infrastructure."
David Schneider
"You have to wonder how the buffer sizing rule of thumb lasted a decade, and further, whether the answer to this question might be that large buffers don't pose as grave a threat as the authors would have us believe.”
Justin Costa-Roberts
7
8
The Story
10,000
20
# packets
at 10Gb/s
1,000,000
E