From: Gleb Natapov <gleb@redhat.com>
To: Victor fernandez <Victormanuel.fernandez@usc.es>
Cc: kvm@vger.kernel.org
Subject: Re: Lost of performance over AMD-V with KVM?
Date: Tue, 9 Jul 2013 17:47:22 +0300 [thread overview]
Message-ID: <20130709144722.GD24941@redhat.com> (raw)
In-Reply-To: <51DC1EC4.4000001@usc.es>
On Tue, Jul 09, 2013 at 04:31:32PM +0200, Victor fernandez wrote:
> Hi Gleb,
>
> there are flags that with the tuning of KVM (-cpu host) we get
> that some of
> important cpu flags are not included in SL6 ( Ht, rdtscp,
> 3dnowext, 3dnow,
> constant_tsc,nonstop_tsc,amd_dcm,monitor,svm,extapic,osvw,ibs,skinit,wdt,nodeid_msr).
>
Not all of them will be passed to a guest even with -cpu host, but
instruction set related once like 3dnowext/3dnow you should be.
What is qemu command line that you use?
> And the machines used for testing were 2 nodes AMD 6128 - 2Ghz -
> 2 CPUs
> (8 Cores/CPU Core - 1GB) 16 Cores Total - 16 GB RAM for AMD-V
> virtualization,
> and we did the same tests with the same software versions with 2
> nodes of DELL
> PowerEdge SC1950, biprocessors dual core, 2 Xeon(R) CPU 5160 @
> 3.00GHz,
> 16GB RAM. for Intel VT virtualization. And with Intel technology
> we get the
> typical virtualization performance, around 2-5% of lost.
>
> Thanks in advance,
> Víctor Fdez.
>
> El 09/07/13 15:33, Gleb Natapov escribió:
> >On Tue, Jul 09, 2013 at 02:20:43PM +0200, Victor fernandez wrote:
> >>Hi EveryBody!
> >>
> >>recently, we have tested some specific High Energy Physics benchmarks that
> >>was running over S.L.6 and we observed that the performance over AMD
> >>machines was really bad, about 30-40% lost. We were investigating
> >>and we conclude that the problem was with some cpu flags. These cpu
> >>flags are important for running our application because it is optimazed
> >>for running with them. So, we tuning our KVM with the name of flags
> >>parameters and with "-cpu host" parameter too, but
> >>we observed that the flags that we needed, they weren't included when
> >>we started the S.L.6 instance. We try to solve it, but when we conclude
> >>that the specific flags weren't included in the code of the last
> >>versions of KVM
> >>and LibVirt(capabilities). Do you know something about this problem or
> >>you are aware of this?
> >>
> >What flags are missing?
> >
> >--
> > Gleb.
>
--
Gleb.
next prev parent reply other threads:[~2013-07-09 14:47 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2013-07-09 12:20 Lost of performance over AMD-V with KVM? Victor fernandez
2013-07-09 13:33 ` Gleb Natapov
[not found] ` <51DC1EC4.4000001@usc.es>
2013-07-09 14:47 ` Gleb Natapov [this message]
[not found] ` <51DC24C9.8080700@usc.es>
2013-07-09 15:00 ` Gleb Natapov
[not found] ` <51DC274D.8090409@usc.es>
2013-07-09 15:16 ` Gleb Natapov
[not found] ` <51DC2C92.4020605@usc.es>
2013-07-09 16:08 ` Gleb Natapov
2013-07-11 12:52 ` Victor fernandez
2013-07-11 14:07 ` Gleb Natapov
[not found] ` <51DEC106.4040300@usc.es>
2013-07-11 14:43 ` Gleb Natapov
2013-07-09 15:17 ` Victor fernandez
2013-07-09 14:42 ` Victor fernandez
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20130709144722.GD24941@redhat.com \
--to=gleb@redhat.com \
--cc=Victormanuel.fernandez@usc.es \
--cc=kvm@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.