[partially solved] tcp_window_scaling degrades performance

[partially solved] tcp_window_scaling degrades performance

[Maciej Soltysiak]

> AK> It is very very likely the firewall, window scaling works for a lot
> AK> of people.
> It is probable, but here: only 2.6.7+ machines behave like this.
> I also noticed, that turning tcp_window_scaling off does not always
> fix the problem, turning tcp_bic to 0 too helps even more.
It appears that checkpoint fw-1 that is here spoils everything so that
current linux boxes need to disable tcp_windows_scaling to get
reasonable throughput. When I switched a server to a different link
that bypasses checkpoint it worked well.

Although it most propably is checkpoint's fault (This one is based on
linux 2.4.9-39cpsmp) I belive that the change in the kernel that started
producing these problems is this one:
http://linux.bkbits.net:8080/linux-2.6/cset@1.1784.10.7?nav=index.html|ChangeSet@-3w

When I removed this patch from a 2.6.8-rc1 kernel it started to work
good again. But of course it may be a blind shot.

I am planning on investigating this issue from the checkpoint's
perspective.

Anyway if there is anyone willing to investigate the tcpdumps and
tcptraces of the slow/fast throughput, here it is:
http://soltysiak.com/tcp.php

I am no guru, but it shows:
1) slower throughput
2) twice as much packets exchanged
3) different advertised windows

Best regards,
Maciek

re: [partially solved] tcp_window_scaling degrades performance

[Dan Kegel]

Maciej Soltysiak <solt@dns.toxicfilms.tv> wrote:
> It appears that checkpoint fw-1 that is here spoils everything so that
> current linux boxes need to disable tcp_windows_scaling to get
> reasonable throughput.

Just checking- have you read
http://lwn.net/Articles/91976/
("TCP window scaling and broken routers")?

- Dan

-- 
My technical stuff: http://kegel.com
My politics: see http://www.misleader.org for examples of why I'm for regime change