[Qemu-devel] Re: Qemu-devel Digest, Vol 22, Issue 46

[Qemu-devel] Re: Qemu-devel Digest, Vol 22, Issue 46

[Julian Chesterfield]

On Fri, Jan 21, 2005 at 03:52:43PM +0000, Julian Chesterfield wrote:

>> My understanding is that Qemu does not support hyperthreading in the 
>host 
>> OS.
>
>
>You mean guest OS, right? 

Actually I was thinking about the host!

>It shouldn't make a different if the host OS 
>uses it
>or not.
>

Excuse my ignorance but is this because there is no 
potential for parallelising tasks in qemu to take advantage of idle 
CPU cycles?

- Julian

Re: [Qemu-devel] Re: Qemu-devel Digest, Vol 22, Issue 46

[Jim C. Brown]

On Fri, Jan 21, 2005 at 08:53:55PM +0000, Julian Chesterfield wrote:
> 
> 
> On Fri, Jan 21, 2005 at 03:52:43PM +0000, Julian Chesterfield wrote:
> 
> >> My understanding is that Qemu does not support hyperthreading in the 
> >host 
> >> OS.
> >
> >
> >You mean guest OS, right? 
> 
> Actually I was thinking about the host!

Why???

Hyperthreading, or even true SMP, has no effect on qemu whatsoever.

> 
> >It shouldn't make a different if the host OS 
> >uses it
> >or not.
> >
> 
> Excuse my ignorance but is this because there is no 
> potential for parallelising tasks in qemu to take advantage of idle 
> CPU cycles?

I don't know about "potential" ... it is possible, but afaik it isn't done
currently in qemu.

> 
> - Julian
> 
> 
> 
> _______________________________________________
> Qemu-devel mailing list
> Qemu-devel@nongnu.org
> http://lists.nongnu.org/mailman/listinfo/qemu-devel
> 

-- 
Infinite complexity begets infinite beauty.
Infinite precision begets infinite perfection.

Re: [Qemu-devel] Re: Qemu-devel Digest, Vol 22, Issue 46

[Paul Brook]

On Friday 21 January 2005 20:53, Julian Chesterfield wrote:
> >It shouldn't make a different if the host OS
> >uses it
> >or not.
>
> Excuse my ignorance but is this because there is no
> potential for parallelising tasks in qemu to take advantage of idle
> CPU cycles?

In theory you could put CPU and system emulation in separate threads. In 
practice I suspect the synchronisation overhead outweighs the (small) benefit 
that hyperthreading gives you. I'm not even sure it's give any significant 
benefit on real SMP systems.

Simulating multiple cpu SMP guest system on a SMP host cpus would be 
interesting, but that's a whole new thing.

Paul

Re: [Qemu-devel] Re: Qemu-devel Digest, Vol 22, Issue 46

[Fabrice Bellard]

Paul Brook wrote:

> Simulating multiple cpu SMP guest system on a SMP host cpus would be 
> interesting, but that's a whole new thing.

I plan to add it someday, but x86 virtualization seems more important 
now. For SMP, some work is needed to render the virtual CPU core thread 
safe (I would like that each virtual CPU use the same translated code 
cache).

Fabrice.