Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

Running FC release 12. Im alaways seeing ksoftirqd/x (x being 0-9) at the top of the processlist, with 100% cpu.

The server has a bonded 2gbit connection, serving files from an SSD array. Currently its using 1.6gbit. Server load is ~ 1.5 (dual quad core). iowait is non-existent.

share|improve this question
    
Which NICs are you using? –  d34dh0r53 Apr 18 '10 at 6:49

2 Answers 2

up vote 1 down vote accepted

This has been a longstanding linux problem. If you do a google search, the reports of this go way back to 2004, without any real answers. This seems to be mostly happening during high network activity, and looks like this is your case. Linux tcp stack has never been the greatest, so I am not surprised. This may also be related to your network interfaces and the modules they are using.

I would suggest trying different network hardware. Maybe shell out for a 10Gbit card if 1Gbit is not enough? Try without bonding?

share|improve this answer
    
"Linux tcp stack has never been the greatest" — what a crap statement, dude. There's no greatest at all, actually, but at least it's been re-written from scratch comparing to BSDes and had things like separate routing tables for years. Linux TCP stack itself is cutting-edge implementation, lots of congestion modules, tunings (not requiring to reboot as often with BSDs), etc. And it never had such silly bugs BSD had (I can mention at least one severe I had found by myself). What's your facts now? –  poige Apr 23 '12 at 12:45

Look at making sure that your nic supports MSI-X (lspci -vv).

You can read more here http://www.alexonlinux.com/msi-x-the-right-way-to-spread-interrupt-load

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.