[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [Xen-users] Does faster 2D Host video card help HVM guest?


  • To: "Trolle Selander" <trolle.selander@xxxxxxxxx>, xen-users@xxxxxxxxxxxxxxxxxxx
  • From: "Petersson, Mats" <Mats.Petersson@xxxxxxx>
  • Date: Mon, 4 Jun 2007 12:06:48 +0200
  • Delivery-date: Mon, 04 Jun 2007 03:05:15 -0700
  • List-id: Xen user discussion <xen-users.lists.xensource.com>
  • Thread-index: AcemjvUnYNSf7TmgTo6uav97PAJLPwAAH3PQ
  • Thread-topic: [Xen-users] Does faster 2D Host video card help HVM guest?

 

> -----Original Message-----
> From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx 
> [mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of 
> Trolle Selander
> Sent: 04 June 2007 10:58
> To: xen-users@xxxxxxxxxxxxxxxxxxx
> Subject: Re: [Xen-users] Does faster 2D Host video card help 
> HVM guest?
> 
> I recently measured graphics performance on an OS/2 guest 
> (which is as "unaccellerated" as it can get) and discovered, 
> much to my surprise, that graphics performance is nowhere 
> near as bad as one might assume, and in fact pretty good. I 
> believe the explanation to be that though unacellerated, the 
> virtual graphics card is sitting on a "bus" that actually has 
> the full speed of RAM, which on a fast dual channel DDR2 
> system is actually faster than even PCIe 16x. Moroever, due 
> to the fact that the cirrus device model uses buffered mmio, 
> in a multi-core system, a certain amount of the "work" can 
> get handed off to dom0 and the other CPU core. 
> It would seems to me that the main reason 2D & gui ops feel 
> "slow" in guests right now is the fact that even with SDL, 
> the guest VM's screen is only updated  at 30 Hz.

Yes, the "slowness" in graphics is more about update frequency than to
do with the actual "read/write" performance of the "graphics adapter".
The way it works (for those inclined to wanting to know these things) is
that the frame-buffer is shared between qemu-dm and the guest. So data
can be read/written with normal RAM-speed, as Trolle says. To preserve
CPU-performance on qemu-dm, updates are rate-limited, because the method
to update the screen is that we read the frame-buffer, and compare it
with an "old" frame buffer [I believe it's actually a check-sum, rather
than byte by byte compare (ok, technically, it's also using SSE
instructions to get as good a throughput as possible)]. Doing this at
50-100Hz would consume a lot more than the current CPU consumption of
qemu-dm. 

--
Mats
> 
> /Trolle
> 
> 
> On 6/4/07, Mark Williamson <mark.williamson@xxxxxxxxxxxx> wrote:
> 
>       > Because of how the qemu-dm graphics card emulation 
> works, I believe host
>       > graphics card speed has pretty much no impact 
> whatsoever with the current
>       > device model. The important factors are host CPU & RAM speed. 
>       
>       Right now, the guest's graphics performance is going to 
> be fairly slow
>       whatever you're running in the host.  It's not worth 
> trying to upgrade the
>       host graphics card, because it's not really going to 
> make any difference to 
>       the guest.
>       
>       Guest display improvements will probably come from 
> future optimisations in
>       future releases of Xen.
>       
>       Cheers,
>       Mark
>       
>       --
>       Dave: Just a question. What use is a unicyle with no 
> seat?  And no pedals! 
>       Mark: To answer a question with a question: What use is 
> a skateboard?
>       Dave: Skateboards have wheels.
>       Mark: My wheel has a wheel!
>       
> 
> 
> 



_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.