FAST TCP Anwis Das Ajay Gulati Slides adapted from : IETF presentation slides Link: .
-
Upload
irving-lomax -
Category
Documents
-
view
225 -
download
1
Transcript of FAST TCP Anwis Das Ajay Gulati Slides adapted from : IETF presentation slides Link: .
FAST TCP
Anwis DasAjay Gulati
Slides adapted from : IETF presentation slidesLink: http://netlab.caltech.edu/FAST/index.html
Background
Last time we saw reliable communication requires a transport level protocol
But user’s are selfish, leading to congestion
Van Jacobson solved the problem by adapting TCP Slow Start Congestion Avoidance Accurate Estimation of RTT, etc
Is the solution still valid considering today’s technology??
Hypothesis
TCP Reno is a good solution for low speed networks but not a viable solution for high-speed networks Too conservative, not stable, and requires
extremely small equilibrium loss probability There exists an equation that governs
the flow of all variants of TCP FAST provides a solution to this equation and
shows that it is stable and reaches equilibrium quickly
Philosophy of Reno
Packet level Designed and implemented first
Flow level Understood afterwards
Flow level dynamics determines Equilibrium: performance, fairness Stability
Design flow level equilibrium & stability Implement flow level goals at packet level
Problems with TCP Reno
Equilibrium problem Packet level: AI too slow, MI too drastic Flow level: required loss probability too
small Dynamic problem
Packet level: must oscillate on binary signal
Flow level: unstable at large window
TCP Variant Performance
ACK: W W + 1/W
Loss: W W – 0.5W
Reno: AIMD (1, 0.5)
ACK: W W + a(w)/W
Loss: W W – b(w)W
ACK: W W + 0.01
Loss: W W – 0.125W
HSTCP: AIMD (a(w), b(w))
STCP: MIMD (1/100, 1/8)
Flow level: Reno, HSTCP, STCP, FAST
Different gain and utility Ui
They determine equilibrium and stability
Different congestion measure pi Loss probability (Reno, HSTCP, STCP) Queueing delay (Vegas, FAST)
Generic flow level dynamics!
windowadjustment
controlgain
flow levelgoal=
FAST TCP Dynamics
Generic equation:
windowadjustment
controlgain
flow levelgoal=
FAST TCP equation:
Can map generic equation to FAST TCP equation and understand how FAST TCP stabilizes faster than other TCP variants
Window control algorithm
Full utilization regardless of bandwidth-delay product
Globally stable exponential convergence
Fairness weighted proportional fairness parameter
Window control algorithm
Theorem (Jin, Wei, Low ‘03) In absence of delay Mapping from w(t) to w(t+1) is contraction Global exponential convergence Full utilization after finite time Utility function: i log xi (proportional fairness)
FAST Linux
throughput
loss
queue
STCPHSTCP
Dynamic sharing on Dummynet capacity = 800Mbps delay=120ms 14 flows iperf throughput Linux 2.4.x (HSTCP: UCL)
30min
Aggregate throughput
small window800pkts
largewindow
8000
Dummynet: cap = 800Mbps; delay = 50-200ms; #flows = 1-14; 29 expts
Fairness
Jain’s index
HST
CP ~
Ren
oDummynet: cap = 800Mbps; delay = 50-200ms; #flows = 1-14; 29 expts
Stability
Dummynet: cap = 800Mbps; delay = 50-200ms; #flows = 1-14; 29 expts
stable indiverse
scenarios
Open issues
1 BaseRTT estimation route changes, dynamic sharing does not upset stability
2 Small network buffer at least like TCP adapt on slow timescale, but how?
3 TCP-friendliness friendly at least at small window tunable, but how to tune?
4 Reverse path congestion should react? rare for large transfer?
5 Did not compare to Vegas Future plans
What people on the street are saying about FAST TCP…
PC Magazine“The fastest hardware in the world will not do you much good if your software can't take advantage of it.Business Week “Connections that leave broadband in the dust” Steven Low “This is OK for driving slowly in a parking lot. But on the open road you need to be able to look further ahead: That's what we are doing with Fast TCP.” Future Zone “Scientists have developed a new data transfer protocol for the Internet fast enough to download a full-length DVD movie in less than five seconds”