FAST II: Algorithms and Performance

1
FAST II Algorithms and Performance

• Cheng Jin
• David Wei
• Steven Low
• http//netlab.caltech.edu

2
Topics

• Design Issues
• FAST (sc2002 version)
• Window Pacing
• Experimental Results

3
Design Issues

• Raj Jains Three questions that a congestion
  control algorithm should answer
• Decision Frequency (Timescale)
• Decision function
• How to Increase/Decrease Window

4
How Frequent?

• Timescale of a congestion control algorithm
• New Reno

5
Decision Function
6
Decision Function using Binary Signal
7
Congestion Signal

• Loss
• Delay
• ECN

8
Increase/Decrease Function

• Under synchronization assumption (Chui and Jain)

9
Congestion Control Algorithms
10
FAST (sc2002) Algorithm

• Do the following every other RTT
• per acknowledgment
• Calculation
• Action
• If w(t)gtw w(t)w(t)-1
• If w(t)ltw w(t)w(t)1

11
FAST (sc2002) Algorithm

• Do the following every other RTT
• per acknowledgment
• Calculation
• Action
• If w(t)gtw w(t)w(t)-1
• If w(t)ltw w(t)w(t)1

12
FAST (sc2002) Algorithm

• Do the following every other RTT
• per acknowledgment
• Calculation
• Action
• If w(t)gtw w(t)w(t)-1
• If w(t)ltw w(t)w(t)1

13
FAST (sc2002) Algorithm

• Do the following every other RTT
• per acknowledgment
• Calculation
• Action
• If w(t)gtw w(t)w(t)-1
• If w(t)ltw w(t)w(t)1

14
FAST (sc2002) Algorithm

• Do the following every other RTT
• per acknowledgment
• Calculation
• Action
• If w(t)gtw w(t)w(t)-1
• If w(t)ltw w(t)w(t)1

15
FAST (sc2002) Algorithm

• Do the following every other RTT
• per acknowledgment
• Calculation
• Action
• If w(t)gtw w(t)w(t)-1
• If w(t)ltw w(t)w(t)1

No more aggressive than Rate Halving
No more aggressive than Slow Start
16
Results on Convergence

• Under synchronization model
• Single link identical sources converge
• Single link heterogeneous sources bounded by
  converging sequences
• General case
• Contraction mapping

17
Simulation with Matlab
pkt/ms
pkt/ms
pkt
pkt

• baseRTT 550ms,600ms,650ms,700ms,750ms
• Alpha500 pkt
• C5000/12 pkt/ms

18
Simulation with Matlab
ms
pkt

• baseRTT 550ms,600ms,650ms,700ms,750ms
• Alpha500
• C5000/12 pkt/ms

19
Window Pacing
20
Pacing

• Why Pacing?
• Eliminate noise of packet loss due to
  burstiness (for loss-based algorithms).
• Eliminate noise of queuing delay due to
  burstiness (for delay-based algorithms).
• Pacing can make the behavior of TCP more
  fluid-like
• Why not pacing?

21
Window Pacing

• Over one RTT

• Action (each acknowledgment)
• If w(t)gtw w(t)w(t)-1
• If w(t)ltw w(t)w(t)1

22
Time-Based Window Pacing
cwnd increments are scheduled at fixed
intervals.
data
data
ack
k
k
k
k
k
k
(Cheng Jin, Caltech)
23
Event-Based Pacing
Detect sufficiently large gap between consecutive
bursts and delay cwnd increment until the end of
each such burst.
(Cheng Jin, Caltech)
24
Implementing Pacing

• Using Intel x86 cycle counter for high
  resolution
• Combination of time-based and event-based pacing
• Time-based pacing is supported only for single
  TCP stream in a single machine
• Event-based pacing can support multiple TCP
  streams in a single machine

(Cheng Jin, Caltech)
25
Experiments

• SC2002
• Dummynet
• Internet2
• Summary

26
SC2002 Network
OC48
OC192
(Sylvain Ravot, Caltech)
27
FAST BMPS
10
flows
9
Geneva-Sunnyvale
7
FAST
2
Baltimore-Sunnyvale

• FAST
• Standard MTU
• Throughput averaged over gt 1hr

1
Internet2 Land Speed Record
2
1
(Steven Low, Caltech)
28
Dummynet Testbed_at_ Netlab, Caltech
29
Static throughput

• Dynamic sharing on Dummynet
• capacity 800Mbps
• Delay 50 - 200ms
• flows 1 - 10
• iperf throughput
• Linux 2.4 (HSTCP UCL)

30
Static throughput
31
Dynamic Sharing
HSTCP
FAST

• Fairness Problem

Scalable
Linux
32
queue
HSTCP
FAST
Linux
loss
throughput
Linux
Scalable
STCP
HSTCP
HSTCP
33
Dynamic Sharing
HSTCP
FAST
Scalable
Linux
34
Abilene Testing
U Washington
PSC
Caltech
GaTech
35
Abilene Testing

• 100Mbps Test Caltech-gtGatech
• Duration 120 second
• Base RTT 46ms
• UDP blast every 20 seconds for 5 seconds per
  blast

RTT (us)
Cwnd
(Cheng Jin, Caltech)
36
Performance Metrics

• Aggregate Average Goodput
• Fairness Jains index
• Stability
• Responsiveness

37
Aggregate throughput

• Dummynet cap 800Mbps delay 50-200ms flows
  1-14 29 expts

38
Fairness

• Dummynet cap 800Mbps delay 50-200ms flows
  1-14 29 expts

39
Stability

• Dummynet cap 800Mbps delay 50-200ms flows
  1-14 29 expts

40
Responsiveness
Dummynet cap 800Mbps delay 50-200ms flows
1-14 29 expts
41
Future Works

• Congestion Window Control Algorithm Stability
  with delay
• Synchronization Model
• Burstiness Control Modeling, Measurement
• Testing Experiments in production networks

42
Questions?

• Thanks?