* Performance numbers
@ 2012-02-06 13:01 Richard Purdie
2012-02-06 13:28 ` Eric Bénard
2012-02-06 15:28 ` Khem Raj
0 siblings, 2 replies; 6+ messages in thread
From: Richard Purdie @ 2012-02-06 13:01 UTC (permalink / raw)
To: openembedded-core
Its been a while since we had some performance numbers and there were
recently some questions about how current build time compared to
previous numbers.
On my 12 core system, build time for qemux86 from scratch is ~42
minutes. This compares favourably with previous numbers from that
system. Interestingly, switching to qemumips resulted in a build time of
43 minutes in the same tmpdir suggesting that the native part of the
bootstrap process isn't the bottleneck.
On the quad core (the system used for many of the previous numbers I'd
provided), build time for qemux86 was 105 minutes which again is roughly
consistent with where things were for 1.1.
So things haven't changed much recently which is as I'd expect. We
haven't seen many performance related patches but its good to know
things aren't regressing either.
Cheers,
Richard
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: Performance numbers
2012-02-06 13:01 Performance numbers Richard Purdie
@ 2012-02-06 13:28 ` Eric Bénard
2012-02-06 13:32 ` Martin Jansa
2012-02-06 15:28 ` Khem Raj
1 sibling, 1 reply; 6+ messages in thread
From: Eric Bénard @ 2012-02-06 13:28 UTC (permalink / raw)
To: openembedded-core
Hi Richard,
Le Mon, 06 Feb 2012 13:01:15 +0000,
Richard Purdie <richard.purdie@linuxfoundation.org> a écrit :
> Its been a while since we had some performance numbers and there were
> recently some questions about how current build time compared to
> previous numbers.
>
> On my 12 core system, build time for qemux86 from scratch is ~42
> minutes. This compares favourably with previous numbers from that
> system. Interestingly, switching to qemumips resulted in a build time of
> 43 minutes in the same tmpdir suggesting that the native part of the
> bootstrap process isn't the bottleneck.
>
> On the quad core (the system used for many of the previous numbers I'd
> provided), build time for qemux86 was 105 minutes which again is roughly
> consistent with where things were for 1.1.
>
what are your settings (PARALLEL_MAKE & BB_NUMBER_THREADS) used in
both cases ?
Thanks
Eric
--
http://eukrea.com/en/news/104-2012
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: Performance numbers
2012-02-06 13:28 ` Eric Bénard
@ 2012-02-06 13:32 ` Martin Jansa
2012-02-06 13:57 ` Richard Purdie
0 siblings, 1 reply; 6+ messages in thread
From: Martin Jansa @ 2012-02-06 13:32 UTC (permalink / raw)
To: Patches and discussions about the oe-core layer
[-- Attachment #1: Type: text/plain, Size: 1210 bytes --]
On Mon, Feb 06, 2012 at 02:28:30PM +0100, Eric Bénard wrote:
> Hi Richard,
>
> Le Mon, 06 Feb 2012 13:01:15 +0000,
> Richard Purdie <richard.purdie@linuxfoundation.org> a écrit :
>
> > Its been a while since we had some performance numbers and there were
> > recently some questions about how current build time compared to
> > previous numbers.
> >
> > On my 12 core system, build time for qemux86 from scratch is ~42
> > minutes. This compares favourably with previous numbers from that
> > system. Interestingly, switching to qemumips resulted in a build time of
> > 43 minutes in the same tmpdir suggesting that the native part of the
> > bootstrap process isn't the bottleneck.
> >
> > On the quad core (the system used for many of the previous numbers I'd
> > provided), build time for qemux86 was 105 minutes which again is roughly
> > consistent with where things were for 1.1.
> >
> what are your settings (PARALLEL_MAKE & BB_NUMBER_THREADS) used in
> both cases ?
And what's the target in your tests?
My quad core does core-image-minimal-mtdutils for qemux86* from scratch
in ~ 100 minutes.
Cheers,
--
Martin 'JaMa' Jansa jabber: Martin.Jansa@gmail.com
[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 205 bytes --]
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: Performance numbers
2012-02-06 13:32 ` Martin Jansa
@ 2012-02-06 13:57 ` Richard Purdie
0 siblings, 0 replies; 6+ messages in thread
From: Richard Purdie @ 2012-02-06 13:57 UTC (permalink / raw)
To: Patches and discussions about the oe-core layer
On Mon, 2012-02-06 at 14:32 +0100, Martin Jansa wrote:
> On Mon, Feb 06, 2012 at 02:28:30PM +0100, Eric Bénard wrote:
> > Hi Richard,
> >
> > Le Mon, 06 Feb 2012 13:01:15 +0000,
> > Richard Purdie <richard.purdie@linuxfoundation.org> a écrit :
> >
> > > Its been a while since we had some performance numbers and there were
> > > recently some questions about how current build time compared to
> > > previous numbers.
> > >
> > > On my 12 core system, build time for qemux86 from scratch is ~42
> > > minutes. This compares favourably with previous numbers from that
> > > system. Interestingly, switching to qemumips resulted in a build time of
> > > 43 minutes in the same tmpdir suggesting that the native part of the
> > > bootstrap process isn't the bottleneck.
> > >
> > > On the quad core (the system used for many of the previous numbers I'd
> > > provided), build time for qemux86 was 105 minutes which again is roughly
> > > consistent with where things were for 1.1.
> > >
> > what are your settings (PARALLEL_MAKE & BB_NUMBER_THREADS) used in
> > both cases ?
>
> And what's the target in your tests?
>
> My quad core does core-image-minimal-mtdutils for qemux86* from scratch
> in ~ 100 minutes.
Sorry, I should have said. Both are for building core-image-sato which
is what I usually use as the benchmark. The quad core is running 8 bb
threads and 4 parallel make. The 12 core was set to 48/48 but anything
over 24 gives the build time above and this is effectively "unlimited"
as it very rarely hits 48 tasks in parallel.
Cheers,
Richard
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: Performance numbers
2012-02-06 13:01 Performance numbers Richard Purdie
2012-02-06 13:28 ` Eric Bénard
@ 2012-02-06 15:28 ` Khem Raj
2012-02-06 16:22 ` Richard Purdie
1 sibling, 1 reply; 6+ messages in thread
From: Khem Raj @ 2012-02-06 15:28 UTC (permalink / raw)
To: Patches and discussions about the oe-core layer
On Mon, Feb 6, 2012 at 5:01 AM, Richard Purdie
<richard.purdie@linuxfoundation.org> wrote:
>
> On my 12 core system, build time for qemux86 from scratch is ~42
> minutes. This compares favourably with previous numbers from that
> system. Interestingly, switching to qemumips resulted in a build time of
> 43 minutes in the same tmpdir suggesting that the native part of the
> bootstrap process isn't the bottleneck.
are these both build from scratch or did mips use the native parts
previously generated by x86
I have also seen mips take a bit of more time. I think it could be
owed to toolchain.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: Performance numbers
2012-02-06 15:28 ` Khem Raj
@ 2012-02-06 16:22 ` Richard Purdie
0 siblings, 0 replies; 6+ messages in thread
From: Richard Purdie @ 2012-02-06 16:22 UTC (permalink / raw)
To: Patches and discussions about the oe-core layer
On Mon, 2012-02-06 at 07:28 -0800, Khem Raj wrote:
> On Mon, Feb 6, 2012 at 5:01 AM, Richard Purdie
> <richard.purdie@linuxfoundation.org> wrote:
> >
> > On my 12 core system, build time for qemux86 from scratch is ~42
> > minutes. This compares favourably with previous numbers from that
> > system. Interestingly, switching to qemumips resulted in a build time of
> > 43 minutes in the same tmpdir suggesting that the native part of the
> > bootstrap process isn't the bottleneck.
>
> are these both build from scratch or did mips use the native parts
> previously generated by x86
> I have also seen mips take a bit of more time. I think it could be
> owed to toolchain.
It used the native bits already generated so yes, it could be toolchain
speed differences...
Cheers,
Richard
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2012-02-06 16:30 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2012-02-06 13:01 Performance numbers Richard Purdie
2012-02-06 13:28 ` Eric Bénard
2012-02-06 13:32 ` Martin Jansa
2012-02-06 13:57 ` Richard Purdie
2012-02-06 15:28 ` Khem Raj
2012-02-06 16:22 ` Richard Purdie
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox