White screen

White screen

[Mark Clarkson]

Hi,
When using WinXP I get a completely white screen whenever I change the
resolution and just before the Windows GUI comes up, which lasts for
around 20 seconds. Also, as with Xen HVM, network performance is poor,
although it seems to be faster than WinXP under Xen. Do you expect
network performance to be good enough for server use in a guest in the
future, maybe once the MMU has been improved, or will you go the same
route as Xen and provide non-free guest network drivers?

Tested with stock linux-2.4.20-rc1 kvm modules and svn modules/qemu
under Debian unstable.

Thanks
Mark.

PS. Highly impressive to have two WinXP windows playing full screen
video in real time, and with plenty of cpu power left!


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV

Re: White screen

[Avi Kivity]

Mark Clarkson wrote:
> Hi,
> When using WinXP I get a completely white screen whenever I change the
> resolution and just before the Windows GUI comes up, which lasts for
> around 20 seconds. 

That is a known issue.  Windows is clearing the screen using mmio 
writes.  We have hacks to convert the mmio writes to regular RAM writes, 
but they aren't effective for the screen clearing.  It is likely 
fixable, just haven't gotten around to it.


> Also, as with Xen HVM, network performance is poor,
> although it seems to be faster than WinXP under Xen. Do you expect
> network performance to be good enough for server use in a guest in the
> future, maybe once the MMU has been improved, or will you go the same
> route as Xen and provide non-free guest network drivers?
>   

We intend, at first, to improve fully virtualized performance as far as 
we are able, without resorting to paravirtualized drivers.

Should performance not be satisfactory to our needs, we may write 
paravirtualized devices and drivers.  If we do so, I can't commit as to 
whether the paravirtualized Windows drivers will be open or not (I'm not 
saying the drivers will be closed -- just that no decision has been made).

However, if you (or anyone) is interested in writing paravirtulized 
drivers, we will work with you towards merging them into the kernel and 
into the kvm distribution.  We welcome such contributions and will 
cooperate fully.


-- 
error compiling committee.c: too many arguments to function


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV

Re: White screen

[Dor Laor]

> Mark Clarkson wrote:
> > Hi,
> > When using WinXP I get a completely white screen whenever I change
the
> > resolution and just before the Windows GUI comes up, which lasts for
> > around 20 seconds.
> 
> That is a known issue.  Windows is clearing the screen using mmio
> writes.  We have hacks to convert the mmio writes to regular RAM
writes,
> but they aren't effective for the screen clearing.  It is likely
> fixable, just haven't gotten around to it.
> 
> 
> > Also, as with Xen HVM, network performance is poor,
> > although it seems to be faster than WinXP under Xen. Do you expect
> > network performance to be good enough for server use in a guest in
the
> > future, maybe once the MMU has been improved, or will you go the
same
> > route as Xen and provide non-free guest network drivers?
> >
> 
> We intend, at first, to improve fully virtualized performance as far
as
> we are able, without resorting to paravirtualized drivers.
> 
> Should performance not be satisfactory to our needs, we may write
> paravirtualized devices and drivers.  If we do so, I can't commit as
to
> whether the paravirtualized Windows drivers will be open or not (I'm
not
> saying the drivers will be closed -- just that no decision has been
made).
> 
> However, if you (or anyone) is interested in writing paravirtulized
> drivers, we will work with you towards merging them into the kernel
and
> into the kvm distribution.  We welcome such contributions and will
> cooperate fully.
> 

Another future option would be to use hardware that support
virtualization.  There are already NICs (www.netxen.com) and HBAs that
provide segmented virtual NICs per VM. They are not wide spread but they
will be in the near future. The NIC vendor will provide the guest driver
and we will add the VMM side.
Until then para-virtualizes drivers will have to be used.

> 
> --
> error compiling committee.c: too many arguments to function
> 
> 
>
------------------------------------------------------------------------
-
> Take Surveys. Earn Cash. Influence the Future of IT
> Join SourceForge.net's Techsay panel and you'll get the chance to
share
> your
> opinions on IT & business topics through brief surveys - and earn cash
>
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDE
V
> _______________________________________________
> kvm-devel mailing list
> kvm-devel-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org
> https://lists.sourceforge.net/lists/listinfo/kvm-devel

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV

Re: White screen

[Mark Clarkson]

On Mon, 2006-12-18 at 10:07 +0200, Avi Kivity wrote:
> We intend, at first, to improve fully virtualized performance as far
> as 
> we are able, without resorting to paravirtualized drivers.
> 
> Should performance not be satisfactory to our needs, we may write 
> paravirtualized devices and drivers.  If we do so, I can't commit as
> to 
> whether the paravirtualized Windows drivers will be open or not (I'm
> not 
> saying the drivers will be closed -- just that no decision has been
> made).
> 
> However, if you (or anyone) is interested in writing paravirtulized 
> drivers, we will work with you towards merging them into the kernel
> and 
> into the kvm distribution.  We welcome such contributions and will 
> cooperate fully. 

Unfortunately the slow network performance and the fact that, on my
setup at least, network io takes 100% of a single core means that kvm is
not currently useful for server use.

I really like the idea of kvm over Xen as it's a much 'cleaner' solution
and would dearly love to see paravirtualised network drivers, at least
for Linux guests, which would allow me to consider moving from Xen to
kvm.

Do you think it possible/probable for someone with around 10 years
knowledge of Linux and C programming but with no kernel module writing
experience to be able to write a paravirtualised network driver for
linux in under 6 months (in his/her spare time)? Is there any example
code around for this? Would it be a case of say, modifying an ne2k
driver? In fact, I've literally just seen that paravirtualised drivers
exist for xen for linux 2.6 kernels and I presume these are open source
- could these be used in some way?


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV

Re: White screen

[Avi Kivity]

Mark Clarkson wrote:
> Unfortunately the slow network performance and the fact that, on my
> setup at least, network io takes 100% of a single core means that kvm is
> not currently useful for server use.
>
>   

Certainly, kvm is slow right now for server use.  This will improve 
quickly though.

> I really like the idea of kvm over Xen as it's a much 'cleaner' solution
> and would dearly love to see paravirtualised network drivers, at least
> for Linux guests, which would allow me to consider moving from Xen to
> kvm.
>
> Do you think it possible/probable for someone with around 10 years
> knowledge of Linux and C programming but with no kernel module writing
> experience to be able to write a paravirtualised network driver for
> linux in under 6 months (in his/her spare time)? 

Well, it depends on how much spare time said someone is willing to 
commit.  This is a fairly complex system.  I will provide guidance 
however, should anyone require it.

For someone who is familiar with Linux and Windows networking, the task 
should take a few weeks.

> Is there any example
> code around for this? Would it be a case of say, modifying an ne2k
> driver? 

Certainly that's the easiest way to start, as only the xmit/receive 
paths need to be accelerated.

> In fact, I've literally just seen that paravirtualised drivers
> exist for xen for linux 2.6 kernels and I presume these are open source
>   

I understand only the Linux guest drivers are open source.

The Xen paravirtualized drivers are quite complex and heavily tied to 
Xen (by using Xen features such as grant tables and event channels, 
which are all unnecessary for kvm).  While we might borrow code from 
them, outright sharing is questionable.

-- 
error compiling committee.c: too many arguments to function


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV