On Thu, Jan 24, 2013 at 09:38:36AM +0100, Romain Francoise wrote:
> Hi Willy,
>
> Willy Tarreau <[email protected]> writes:
>
> > So that means that we have never built RDRAND on 2.6.32/2.6.34 yet then !
>
> No, it just means it can't be disabled; it's enabled by default.
Indeed.
> I have some strange performance issues with v2.6.32.60, I upgraded a bunch
> of virtual machines which were previously running v2.6.32.59 just fine,
> and with the new kernel the same workload is two to ten times slower. I
> can't bisect easily so I tried disabling RDRAND (thinking that maybe it
> got used even though it's not supported, causing vm exits) and noticed
> that the prompt wasn't showing up.
>
> (And as it turns out, even after disabling ARCH_RANDOM the performance
> issues are still there...)
What is your workload ? Is is mostly network processing, user-land
processing, block I/O ? Did you observe an increase of CPU usage or
is it just that the kernel seems to be idling more or waiting for
I/O completion ? If you observe that a certain process is slower,
could you check with strace if it seems to behave similarly ?
There were a bunch of kvm fixes in 2.6.32.60, are you using KVM ?
If it can be of any help, I find it easier to review the changes
by commit message :
git log --pretty=oneline v2.6.32.59..v2.6.32.60^ | sort -k2
If we're now digging for a 2.6.32.y performance regression, I think we
can remove Peter and Paul from Cc in futher communications to avoid
needlessly bothering them.
Regards,
Willy
--
To unsubscribe from this list: send the line "unsubscribe stable" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
