My experience is that when switching CPUs for the IRQ Interrupts, it introduces
a delay.
In other words, each time it switches CPUs, I would lose interrupts.
I am using CentOS for Telephony purposes, so eached dropped interrupts is a
potential problem.
The first thing that I do on any CentOS box, is disable irqbalance and set them
CPUs manually.
--
--
Steven
http://www.glimasoutheast.org
"Steve Snyder" <swsnyder at insightbb.com> wrote in message
news:200608250848.57118.swsnyder at insightbb.com...> I've got several SMP machines, some running CentOS 4.3 and some running
> Fedora Core 4. All machine are kept fully updated. A few are
> Pentium3-based and a few are Pentium4-based. They are all running the
> irqbalance daemon.
>
> The distribution of interrupts across CPUs is indeed kept balanced, yet
> even after months of uptime ps shows no CPU use whatsoever by irqbalance.
> This from a program that is supposed to wake up every 10 (?) seconds and
> examine the interrupt counts. Given that there is an IRQ balancing
> scheme in the kernel, I have to wonder if irqbalance is actually being
> used.
>
> The man page for this program consists of a single-sentence description of
> what the program does. Nice.
>
> Does irqbalance really do anything on contemporary Red Hat-based systems?
>
> Thanks.