On 2007/04/20 15:11, Mitja wrote: > > Do you mean interrupt%? bsd.mp will probably drop that *way* down. > > No, memory management routines (netstat -m).
ah, right. The thing to look at there is the difference between peak and max of mbuf clusters in use. Also look at net.inet.ip.ifq sysctl, if there are some drops, you can bump net.inet.ip.ifq.maxlen gradually until the drops stop. > Are you saying I should try > with bsd.mp kernel with a single core cpu (opteron 146)? If you are spending a lot of cpu time processing interrupts, see top(1), this should make quite a difference. If you are spending nearly 100% processing interrupts even with no network traffic, then either the bios update, or changing to ACPI, would help that. > > You may need to also disable USB (may have been one of the things fixed > > by moving to either acpi or a newer bios, I don't recall). > > Damn....I was speaking too fast, it just locked up again. This time > after 38 hours, it is an improvement, though. > > > Most of mine are on original bios, -current from around the time 4.1 was > > It is a production router so I can't update to current just now, at > least until I get a backup. I'll try to update bios later today. You need this commit: http://marc.info/?l=openbsd-cvs&m=117200696111354&w=2 Updating the bios fixes some possible interrupt problems but won't stop hangs/crashes. > Thank you for you hints. No trouble, I spent 6+ months trying different things on these, if I can save someone else some of the pain that will be good!
