Re: ebtables and bridge performance

On Tuesday 31 December 2002 23:24, me athome wrote:
> I have been testing Linux as a bridge on a dual P4
> (Xeon) system. I need to do packet mangling, mainly
> just send packets from specific interfaces to IMQ
> devices. The system has two Intel GigE interfaces
> (optical) and I use VLANs to create sub interfaces.
> When using kernel 2.4.20 I get about 160Mbps (highly
> depended on the specific packet size).
>
> I have two related questions:
> 1) After applying:
> ebtables-v2.0.003_vs_2.4.20.diff.txt and
> bridge-nf-0.0.10-against-2.4.20.diff.txt the
> performance under the same exact conditions drops to
> under 100Mbps (without any actual filtering
> configured). Why?

I would be strongly surprised if empty ebtables tables would do that to your 
performance. Have you tested the performance with a kernel without applying 
the bridge-nf patch, so only the ebtables patch?

> 2) Is it possible to configure the bridge such that
> will either use or not use the bridge filtering code?
> Since not all the bridges will need to be filtered I
> would like to at least retain the performance on those
> bridges that do not need to be filtered.

No, that's currently impossible.

cheers,
Bart



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf