running Windows (albeit slowly)

running Windows (albeit slowly)

[Eric S. Johansson]

forgive me if this has been beaten to death but I've been thinking again 
about the problem of running Windows as a guest OS under xen.  I 
understand and accept all the arguments of why you need to modify the 
core OS for your virtualization process.

Under the opinion that there is no crime if optimization fails on 
incorrect code, would it not be appropriate to do all the things 
necessary to run Windows unmodified albeit poorly slash slowly?
assuming of course that people want to run Windows.  The only reason I 
do is because of speech recognition requirements but that's a very 
different conversation.

of course, Linux would continue to run at its normal speed.

---eric

-- 
"Part of the problem with the Wal-Mart business model is that it
requires more poverty in order to grow."

http://www.salon.com/mwt/feature/2004/11/22/wal_mart/print.html


-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now. 
http://productguide.itmanagersjournal.com/

Re: running Windows (albeit slowly)

[Mark Williamson]

> Under the opinion that there is no crime if optimization fails on incorrect 
> code, would it not be appropriate to do all the things necessary to run 
> Windows unmodified albeit poorly slash slowly?

It's not just that the modifications improve performance, guests actually 
need to be modified to work on Xen at all.

When Intel's VT extensions come out, supporting unmodified guests will be 
rather easier.

If you want a free means of running a Windows virtual machine your best 
bet right now is probably QEmu.  Faster than Bochs although with less 
complete machine emulation.  The author intends to enhance it further to 
eventually get VMWare levels of performance (although that's a non-trivial 
undertaking!).

> assuming of course that people want to run Windows.  The only reason I do is 
> because of speech recognition requirements but that's a very different 
> conversation.

Yeah, speech recognition on Linux (and others) is a bit of a pain. 
There are a few open source products (e.g. sphinx, perlbox) but I doubt 
they're usuable for dictation, if that's what you want.  Out of interest, 
does Wine have any success at running NaturallySpeaking / ViaVoice / 
whatever?

Cheers,
Mark


-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now. 
http://productguide.itmanagersjournal.com/

Re: running Windows (albeit slowly)

[Steven Hand]

> forgive me if this has been beaten to death but I've been thinking again 
> about the problem of running Windows as a guest OS under xen.  I 
> understand and accept all the arguments of why you need to modify the 
> core OS for your virtualization process.
> 
> Under the opinion that there is no crime if optimization fails on 
> incorrect code, would it not be appropriate to do all the things 
> necessary to run Windows unmodified albeit poorly slash slowly?
> assuming of course that people want to run Windows.  The only reason I 
> do is because of speech recognition requirements but that's a very 
> different conversation.

To run unmodified arbitrary OS images you need to be able to 

   a) provide 'shadow mode' page tables and segmentation 
   b) trap and virtualize trappable instructions, 
   c) dynamically scan code for untrappable instructions
      and rewrite. 

We do most of (a) already; doing (b) is not hard but not done. 
Doing (c) is tedious, difficult and fragile. 


We're happy for anyone out there to have a go at this, but
we're unlikely to put any effort into it ourselves at this 
time. There may be hardware support in the future which makes
(a) and (b) easier and (c) disappear; or MS may themselves
produce a version of XP which doesn't require (c). Who knows 
if and when either of these will happen tho...


cheers,

S.


-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now. 
http://productguide.itmanagersjournal.com/

Re: running Windows (albeit slowly)

[Eric S. Johansson]

Mark Williamson wrote:
>> Under the opinion that there is no crime if optimization fails on 
>> incorrect code, would it not be appropriate to do all the things 
>> necessary to run Windows unmodified albeit poorly slash slowly?
> 
> 
> It's not just that the modifications improve performance, guests 
> actually need to be modified to work on Xen at all.

I'm sorry.  I misunderstood what was written.  I was under the 
impression that to run Windows or unmodified operating systems quickly 
required a great deal of complexity on the order of vmware and that you 
could run Windows slower (under certain circumstances) with "simple" 
emulation features

> If you want a free means of running a Windows virtual machine your best 
> bet right now is probably QEmu.  Faster than Bochs although with less 
> complete machine emulation.  The author intends to enhance it further to 
> eventually get VMWare levels of performance (although that's a 
> non-trivial undertaking!).

I was under the impression that the performance degradation for qemu was 
on the order of 3x native which would be a bit problematic for speech 
recognition.  Is this impression wrong as well?


> Yeah, speech recognition on Linux (and others) is a bit of a pain. There 
> are a few open source products (e.g. sphinx, perlbox) but I doubt 
> they're usuable for dictation, if that's what you want.  Out of 
> interest, does Wine have any success at running NaturallySpeaking / 
> ViaVoice / whatever?

tell me about it.  I am heavily involved in the Open Source Speech 
Recognition Initiative (www.ossri.org) and just trying to figure out 
what's necessary to make NaturallySpeaking work on wine is a handful 
because we need to make it tremendous number of assumptions about 
NaturallySpeaking.  Scansoft, the current owner of the desktop speech 
recognition monopoly, views customers as a hostile necessity and doesn't 
provide any support to anybody for any reason unless you show up with a 
big fat checkbook.

The open source speech recognition engines are "cute research projects". 
  To get something useful will take approximately five to eight years of 
work and $40 million.  in other words, we're going to need some major 
grants.

as for wine and NaturallySpeaking, we can't get past install.  Again we 
are going to hunt for grants sometime after the first of the year to pay 
for the evaluation and development.  It wouldn't be entirely fair to go 
to the wine community without some compensation for their efforts.

---eric


-- 
"Part of the problem with the Wal-Mart business model is that it
requires more poverty in order to grow."

http://www.salon.com/mwt/feature/2004/11/22/wal_mart/print.html


-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now. 
http://productguide.itmanagersjournal.com/

Re: running Windows (albeit slowly)

[Dave Feustel]

On Wednesday 24 November 2004 08:09 am, Mark Williamson wrote:
> When Intel's VT extensions come out, supporting unmodified guests will be
> rather easier.

Will VT be a set of fake system registers readable/settable by  Xen and guest 
operating systems? 


-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now. 
http://productguide.itmanagersjournal.com/

Re: running Windows (albeit slowly)

[M.A. Williamson]

> I'm sorry.  I misunderstood what was written.  I was under the 
> impression that to run Windows or unmodified operating systems quickly 
> required a great deal of complexity on the order of vmware and that you 
> could run Windows slower (under certain circumstances) with "simple" 
> emulation features

For virtualisation in general, this statement is true: slower solutions 
(e.g. QEmu) that support unmodified OSs are easier to do than VMWare-style 
virtualisation. For Xen right now you'd be looking at running something 
like this on top of XenLinux in order to run unmodified OSs.

> I was under the impression that the performance degradation for qemu was 
> on the order of 3x native which would be a bit problematic for speech 
> recognition.  Is this impression wrong as well?

No, it's pretty slow at the moment. I imagine you'd have the power to get 
some speech recognition running, even with this performance hit but it 
doesn't seem very sane to burn all of your CPU on it and only get a low 
performance out :-( Also, I just discovered that QEmu doesn't emulate sound 
*input* at the moment, which is also a problem!

> The open source speech recognition engines are "cute research projects". 
>   To get something useful will take approximately five to eight years of 
> work and $40 million.  in other words, we're going to need some major 
> grants.

Good luck with this stuff.  I do miss speech recognition when using Linux...

Cheers,
Mark


-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now. 
http://productguide.itmanagersjournal.com/

Re: running Windows (albeit slowly)

[David Hopwood]

Mark Williamson wrote:
>> Under the opinion that there is no crime if optimization fails on 
>> incorrect code, would it not be appropriate to do all the things 
>> necessary to run Windows unmodified albeit poorly slash slowly?
> 
> It's not just that the modifications improve performance, guests 
> actually need to be modified to work on Xen at all.

Supporting full virtualization on x86 is *much* more complicated
than Xen-style paravirtualization, mainly because the x86 has some
instructions that are not privileged, but reveal state that needs to
be virtualized.

> When Intel's VT extensions come out, supporting unmodified guests will 
> be rather easier.

Is there any information available on what the extensions will be?

> Yeah, speech recognition on Linux (and others) is a bit of a pain.

There was recently a big fanfare about IBM open-sourcing some of their
speech recognition code, but I don't know how much development it needs.

-- 
David Hopwood <david.nospam.hopwood@blueyonder.co.uk>



-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now. 
http://productguide.itmanagersjournal.com/