[vortex] Auto-neg or not?

[phase2@minotaur.host4u.net]

On Thu, 26 Sep 2002, Bogdan Costescu wrote:

> On Wed, 25 Sep 2002 phase2@minotaur.host4u.net wrote:
> 
> > Our engineering team doesn't believe auto negotation should be set. Even 
> > after we gave them tests that showed that by NOT setting auto negotiation 
> > we went from over 10,000 KB/s to < 1,000 KB/s.
> 
> You mean - by forcing some media. Autonegotiation should be on by default.
> Duplex mismatch is a sure way to cripple the connection.

Yeah, sorry.  Both the uplink and the card were forced to 100/full and the 
results were less than pleasant.  Most of the worries come from our 
experience with Sun hardware, which definetly has problems with it's auto 
negotiating.

> 
> > Sep 15 20:00:15 qwburbank01 kernel: NET: 2848 messages suppressed. 
> > Sep 15 20:00:41 qwburbank01 kernel: eth0: Too much work in interrupt, 
> > status 8401. 
> 
> No idea about the first one; possibly just a way to throttle down number 
> of messages by bundling them.
> 

Yeah, it's actually a countermeasure for DoS attacks.  The general concept 
being that you start causing enough net problems for the kernel to 
complain about that it'll send disk usage skyrocketing, so instead it just 
bundles the messages to cut down on the writes to disk.  Unfortunately, it 
doesn't let you know WHICH messages are being generated all.  Bummer ):

> The second one was already discussed many times on this list. Search the 
> archives at http://www.scyld.com/mailman/listinfo/vortex/
> You'll have to decide whether the network traffic is most important for 
> you and increase module parameter "max_interrupt_work" accordingly.
> 
> 

Thanks for this one, I've looked at it and I think we're going to try and 
double it ( 64 ).  These machines are dual AMD 2100 CPUB boxes with 2 GB's 
of ram.  They're high traffic web servers, basically serving everything 
via squid.  If anyone has a better suggestion instead of 64, please let me 
know what I should set it to.


Thanks for the help,

sh