All of lore.kernel.org
 help / color / mirror / Atom feed
* another scheduler problem, on x86_64 only?
@ 2005-07-12 21:22 Andrew Theurer
  0 siblings, 0 replies; 5+ messages in thread
From: Andrew Theurer @ 2005-07-12 21:22 UTC (permalink / raw)
  To: xen-devel

I Noticed that dom0 is only getting about 50% cpu after seeing about 1/2 
the performance on some benchmarks, xen0 vs baremetal.  I just ran a 
quick cpu bound script:

while [ 1 ]; do x=1; done&

..and I get this cpu util:

	total   domains
	------- ----------
 cpu0:  [050.7] d0-0[050.7]
 
 cpu0:  [049.8] d0-0[049.8]
 
 cpu0:  [049.8] d0-0[049.8]
 
 cpu0:  [049.8] d0-0[049.8]


Anyone else seen this on x86_64 dom0 with BVT? 

I do not see this in i386 with BVT.

[root@e326-1 src]# xm dmesg | grep sched
(XEN) Using scheduler: Borrowed Virtual Time (bvt)

[root@e326-1 src]# xm dmesg | grep -i changeset
 Latest ChangeSet: Tue Jul 12 08:19:10 2005 
9f6057761c8f485b1948fdd3308ae10e3ea66ad5


-Andrew

^ permalink raw reply	[flat|nested] 5+ messages in thread

* RE: another scheduler problem, on x86_64 only?
@ 2005-07-13 16:37 Nakajima, Jun
  2005-07-13 17:00 ` Stephan Diestelhorst
  0 siblings, 1 reply; 5+ messages in thread
From: Nakajima, Jun @ 2005-07-13 16:37 UTC (permalink / raw)
  To: Andrew Theurer, xen-devel

Andrew Theurer wrote:
> I Noticed that dom0 is only getting about 50% cpu after seeing about
> 1/2 the performance on some benchmarks, xen0 vs baremetal.  I just
> ran a quick cpu bound script:
> 
> while [ 1 ]; do x=1; done&
> 
> ..and I get this cpu util:
> 
> 	total   domains
> 	------- ----------
>  cpu0:  [050.7] d0-0[050.7]
> 
>  cpu0:  [049.8] d0-0[049.8]
> 
>  cpu0:  [049.8] d0-0[049.8]
> 
>  cpu0:  [049.8] d0-0[049.8]
> 
> 
> Anyone else seen this on x86_64 dom0 with BVT?

I also see this problem. For example, if I run lmbench on dom0, I used
to see almost 100% CPU load, but it's clearly limited to <= 50% (and
performance is bad). This stared to happen when the default scheduler is
switched to sedf. I also tried various parameters with sedf, the results
were basically same, i.e. never got > 50%. To me it looks like the
scheduler is always giving 50% CPU cycles to the idle domain(?)
regardless of the scheduling policy.

> 
> I do not see this in i386 with BVT.

So maybe it assumes that the guest kernels run in ring1.

> 
> [root@e326-1 src]# xm dmesg | grep sched
> (XEN) Using scheduler: Borrowed Virtual Time (bvt)
> 
> [root@e326-1 src]# xm dmesg | grep -i changeset
>  Latest ChangeSet: Tue Jul 12 08:19:10 2005
> 9f6057761c8f485b1948fdd3308ae10e3ea66ad5
> 
> 
> -Andrew

Jun
---
Intel Open Source Technology Center

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: another scheduler problem, on x86_64 only?
  2005-07-13 16:37 another scheduler problem, on x86_64 only? Nakajima, Jun
@ 2005-07-13 17:00 ` Stephan Diestelhorst
  2005-07-13 17:24   ` Andrew Theurer
  0 siblings, 1 reply; 5+ messages in thread
From: Stephan Diestelhorst @ 2005-07-13 17:00 UTC (permalink / raw)
  To: xen-devel; +Cc: Andrew Theurer, Nakajima, Jun

Just for clarification:
   Andrew: Which scheduler did you use? sedf? or bvt?
   Jun: did you try the bvt scheduler too?

Stephan
> Andrew Theurer wrote:
> > I Noticed that dom0 is only getting about 50% cpu after seeing about
> > 1/2 the performance on some benchmarks, xen0 vs baremetal.  I just
> > ran a quick cpu bound script:
> >
> > while [ 1 ]; do x=1; done&
> >
> > ..and I get this cpu util:
> >
> > 	total   domains
> > 	------- ----------
> >  cpu0:  [050.7] d0-0[050.7]
> >
> >  cpu0:  [049.8] d0-0[049.8]
> >
> >  cpu0:  [049.8] d0-0[049.8]
> >
> >  cpu0:  [049.8] d0-0[049.8]
> >
> >
> > Anyone else seen this on x86_64 dom0 with BVT?
>
> I also see this problem. For example, if I run lmbench on dom0, I used
> to see almost 100% CPU load, but it's clearly limited to <= 50% (and
> performance is bad). This stared to happen when the default scheduler is
> switched to sedf. I also tried various parameters with sedf, the results
> were basically same, i.e. never got > 50%. To me it looks like the
> scheduler is always giving 50% CPU cycles to the idle domain(?)
> regardless of the scheduling policy.
>
> > I do not see this in i386 with BVT.
>
> So maybe it assumes that the guest kernels run in ring1.
>
> > [root@e326-1 src]# xm dmesg | grep sched
> > (XEN) Using scheduler: Borrowed Virtual Time (bvt)
> >
> > [root@e326-1 src]# xm dmesg | grep -i changeset
> >  Latest ChangeSet: Tue Jul 12 08:19:10 2005
> > 9f6057761c8f485b1948fdd3308ae10e3ea66ad5
> >
> >
> > -Andrew
>
> Jun
> ---
> Intel Open Source Technology Center
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel

^ permalink raw reply	[flat|nested] 5+ messages in thread

* RE: another scheduler problem, on x86_64 only?
@ 2005-07-13 17:13 Nakajima, Jun
  0 siblings, 0 replies; 5+ messages in thread
From: Nakajima, Jun @ 2005-07-13 17:13 UTC (permalink / raw)
  To: Stephan Diestelhorst, xen-devel; +Cc: Andrew Theurer

Stephan Diestelhorst wrote:
> Just for clarification:
>    Andrew: Which scheduler did you use? sedf? or bvt?
>    Jun: did you try the bvt scheduler too?
> 
Yes, I tried to both (either by rebuilding xen or sched=<> boot
parameter), confirming like: 
(XEN) Using scheduler: Borrowed Virtual Time (bvt)

I see the problem no matter which scheduler I choose. I think it should
be a schduling issue on x86_64 xenlinux because the cutoff over CPU
utilization is very obvious; 50% max in the top command, and top is not
confused because benchmarks run slower. 

Jun
---
Intel Open Source Technology Center

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: another scheduler problem, on x86_64 only?
  2005-07-13 17:00 ` Stephan Diestelhorst
@ 2005-07-13 17:24   ` Andrew Theurer
  0 siblings, 0 replies; 5+ messages in thread
From: Andrew Theurer @ 2005-07-13 17:24 UTC (permalink / raw)
  To: Stephan Diestelhorst; +Cc: xen-devel, Nakajima, Jun

Stephan Diestelhorst wrote:

>Just for clarification:
>   Andrew: Which scheduler did you use? sedf? or bvt?
>  
>
BVT.

I'll try sedf with "xm sedf 0 0 0 0 1 0"

-Andrew

^ permalink raw reply	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2005-07-13 17:24 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2005-07-13 16:37 another scheduler problem, on x86_64 only? Nakajima, Jun
2005-07-13 17:00 ` Stephan Diestelhorst
2005-07-13 17:24   ` Andrew Theurer
  -- strict thread matches above, loose matches on Subject: below --
2005-07-13 17:13 Nakajima, Jun
2005-07-12 21:22 Andrew Theurer

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.