On Thu, Dec 13, 2001 at 07:41:37AM +0100, H�kan Stensby wrote:
> > Sounds like there's either a duplex mismatch somewhere, or your ethernet
> > cards/drivers are broken. Can you try verifying your link settings
> > and/or try with different cards?
>
> Might be so, I have a couple o hours every morning (a couple of hours at
> work before those normal people get here). During the day I'll get my
> hands on another card and try tormorrow morning.
Thanks.
> > ICMP source squench is so 1980's.. :) Seriously, it's deprecated, and
> > you shouldn't see these packets anymore.
>
> Yeah, I know but the effect was just like if it was sent.. better to ask
> then to wonder...
OK, sure :)
> > > when running vmstat 1 on the bridge/fw the context switches are ~30 000,
> > > when the slowdowns are it drops to 100-300 (depending how much other
> > > traffic thats going on right now)
> >
> > Forwarding packets doesn't cause context switches. Stop that tcpdump you
> > are running in the background and retry your vmstat test.
>
> Sorry, I read the wrong line, interrupts / sec that is, my fault.
Hmm. It seems that your network card driver doesn't do interrupt mitigation
(30000 interrupts per second is a _lot_), and you seem to get a lot of smaller
packets. If you have a tulip-based NIC around (a lot of cards are), you
might want to download the NAPI patch plus the tulip NAPI driver from
ftp://robur.slu.se/pub/Linux/net-development/NAPI/
and give it a go. With NAPI I can easily do wire speed fast ethernet
bridging/routing on tinygrams (148 thousand packets per second).
> This bridge/fw is up and running and I just must say, again, that it works
> just fine.
Willing to submit a testimonial.. ? :)
cheers,
Lennert
_______________________________________________
Bridge mailing list
[EMAIL PROTECTED]
http://www.math.leidenuniv.nl/mailman/listinfo/bridge
