[Author Prev][Author Next][Thread Prev][Thread Next][Author Index][Thread Index]

Re: IRQ balancing



Hi,

In case you're wondering what this is about: If you see high CPU usage
on one core on your high bandwidth server (>200 Mbit/s, lots of
connections), it is very likely that only one of your CPU is handling
network interrupts.

Let me add a few things I've learned from fighting with this problem:
irqbalance doesn't help a thing, all it does is rotate smp_affinity, so
our 100% usage moved from one core to another. I also tried different
kernels and various e1000e drivers in hope to get MSI-X working, but it
seems as if our hardware doesn't support it (
http://us1.torservers.net/lspci.txt ).

In the end the only thing that helped was to enable Receive Packet
Steering (RPS), which is in the kernel since 2.6.35, but poorly
documented. I still plan to move all the Tor related posts to a separate
blog on torservers, but for now you can find how to enable RPS here:
http://moblog.wiredwings.com/archives/20100827/Howto-Enable-Receive-Packet-Steering-RPS-on-Linux-2.6.35.html

Moritz

On 19.10.2010 00:10, grarpamp wrote:
> Another links regarding earlier posts on this topic:
> 
> http://www.ntop.org/blog/?p=1
> http://www.alexonlinux.com/smp-affinity-and-proper-interrupt-handling-in-linux
> http://www.alexonlinux.com/why-interrupt-affinity-with-multiple-cores-is-not-such-a-good-thing
> ***********************************************************************
> To unsubscribe, send an e-mail to majordomo@xxxxxxxxxxxxxx with
> unsubscribe or-talk    in the body. http://archives.seul.org/or/talk/
> 
***********************************************************************
To unsubscribe, send an e-mail to majordomo@xxxxxxxxxxxxxx with
unsubscribe or-talk    in the body. http://archives.seul.org/or/talk/