On Fri, May 25, 2007 at 01:16:57PM -0700, Jonathan Lundell wrote:
> On May 24, 2007, at 10:51 PM, Andi Kleen wrote:
>
> >>Do we have a feel for how much performace we're losing on those
> >>systems which _could_ do MSI, but which will end up defaulting
> >>to not using it?
> >
> >At least on 10GB ethernet it is a significant difference; you usually
> >cannot go anywhere near line speed without MSI
> >
> >I suspect it is visible on high performance / multiple GB NICs too.
>
> Why would that be? As the packet rate goes up and NAPI polling kicks
> in, wouldn't MSI make less and less difference?
CPUs are so fast now that we never even get into polling mode.
So MSI makes even more of a difference.
davem and jamal hadi salim were already years ago seeing workloads
(packet rates) where the CPU utilization would peak at packet rates
that were just high enough for NAPI to occasionally be used.
IIRC, Jamal's OLS 2005 or 2006 paper talks about this behavior.
> I like the fact that MSI gives us finer control over CPU affinity
> than many INTx implementations, but that's a different issue.
Yes, I agree.
thanks,
grant
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]
