[Iccrg] Best congestion avoidance algorithm

[adam maxiaodong]

Hi, Alejandro,

I have some lab test result for your reference.

Test topology:

vista                    linux 2.6.15 (tc)                   Linux 2.6.24
(reno, bic, vegas ....)
Client ---------------- WAN emulator ----------------------- Server

Server acts as sender, transfer a 2MB file.

(1) Simulate wireless connection
tc qdisc add dev eth0 root handle 1: netem limit 50000 delay 200ms 50ms drop
3 50% duplicate 0 corrupt 1 recorder 5 25 gap 1
tc qdisc add dev eth1 root handle 1: netem limit 50000 delay 200ms 50ms drop
3 50% duplicate 0 corrupt 1 recorder 5 25 gap 1
TCP Stack        Rate(KBps)
bic                    25.7
cubic                36.8
highspeed         28.4
htcp                  26.8
hybla                24.7
lp                     22.1
scalable            28.6
vegas               16.2
veno                 26.8
westwood         25.7
illinois              32.5
yeah                33.6
reno                 16.1

(2) Simulate US-ASIA
tc qdisc add dev eth0 root handle 1: netem limit 50000 delay 175ms 50ms drop
2 duplicate 0  gap 0
tc qdisc add dev eth0 root handle 1: netem limit 50000 delay 175ms 50ms drop
2 duplicate 0  gap 0
TCP Stack        Rate(KBps)
bic                   19.8
cubic                28.7
highspeed         19.1
htcp                  23.7
hybla                19.7
lp                     16.2
scalable            26.9
vegas               15.5
veno                 22.8
westwood         27.2
illinois              25.1
yeah                21.1
reno                16.4

(3) Simulate US-US
tc qdisc add dev eth0 root handle 1: netem limit 50000 delay 30ms 5ms drop
0.1 duplicate 0  gap 0
tc qdisc add dev eth0 root handle 1: netem limit 50000 delay 30ms 5ms drop
0.1 duplicate 0  gap 0
TCP Stack        Rate(KBps)
bic                   310
cubic               402
highspeed        308
htcp                 254
hybla               303
lp                    263
scalable           325
vegas               195
veno                 292
westwood         298
illinois              326
yeah                332
reno                 272

In my test, vegas doesn't perform better than others, but I guess vegas will
perform better when network path has long queuing.  hope this help!

-Adam

On Tue, Jul 8, 2008 at 6:57 AM, Alejandro Cabrera Obed <aco1967 at gmail.com>
wrote:

> Dear all, I have a network scenario with several Windows XP and Linux
> Debian Etch boxes acting as servers and clients, using typical TCP
> traffic like FTP, HTTP, SSH -among others-.
>
> Our boxes mainly have TCP Reno as the TCP congestion avoidance
> algorithm. But we see that TCP Vegas has a better response to
> congestion because it uses round-trip time measure to control the
> transfer rate, opposite to Reno that uses packet loss. And we see
> there are other congestion avoidance algorithm like New Reno, Bic,
> Cubic, etc.
>
> So, what is the best congestion avoidance algorithm nowadays taking
> into account the response time to congestion situations ??? What do
> you recommend to me ???
>
> Special thanks
>
>
> --
> Alejandro Cabrera Obed
> aco1967 at gmail.com
> www.alejandrocabrera.com.ar
>
> _______________________________________________
> Iccrg mailing list
> Iccrg at cs.ucl.ac.uk
> http://oakham.cs.ucl.ac.uk/mailman/listinfo/iccrg
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://oakham.cs.ucl.ac.uk/pipermail/iccrg/attachments/20080708/444cf503/attachment.html