> -----Original Message-----
> From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx
> [mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of
> Antoine Nivard
> Sent: 08 August 2006 07:26
> To: xen-users@xxxxxxxxxxxxxxxxxxx
> Subject: [Xen-users] How many Windows2003 on VT or pacifica
>
> Dear all,
>
> I would like to know how many Windows2003 I can run on 2 dual
> core 3Ghz
> with more than 16 Go RAM.
> - theory?
> - real life(experience)?
In short: How long is a piece of string?
That are idle, or that are doing something useful?
What application(s) are you planning to run?
How much memory does each instance of Win2K3 need?
How much Disk, Network and CPU utilization does it use?
There's no limit to the number of HVM guests (well, there is, but it's
like 100 or more in 32 bit - if you run 64-bit it's ALMOST unlimited,
but may require passing extra arguments to Xen at boot), the real limit
will be how much CPU processing is needed to emulate the devices used by
the guests, plus the amount of CPU needed to support the guest properly
(with the credit scheduler this will be more stable than with the older
schedulers!) and how much memory you give each guest. And that in turn
will be determined by what you're doing within the guest. It's perfectly
possible to make a dual processor machine without virtualization buckle
at the knees from strain if you give it the right load(s), so, as
virtualization adds load to the processor, you can obviously expect that
the machine can start to cause trouble even with one instance of Win2K3
under the "right" (or wrong) circumstances. [Those circumstances are of
course when the system isn't correctly configured for the number of
users or the type of load it's been given - not enough memory or number
of CPU's for example]
On the other hand, most Windows servers are running 15-30% of the CPU
capacity, and nowhere near any of the other hardware limits, so you
could probably run 2-5 of these in a single server. But there will be
some (sometimes significant) overhead in the virtualization situation -
particularly if there is high levels of hardware emulation involved (in
the current implementation for Windows, ALL hardware is emulated -
network, hard-disk, timers, graphics, keyboard, mouse, etc, etc). So a
very disk-intensive application would have significantly increased CPU
usage over the non-virtual version. Work is in progress to improve this,
but it's quite clear that for the foreseeable future (several years),
there will be some overhead in virtualizing the hardware, even if some
extra hardware features are being added to help the processor deal with
the virtualization of other hardware components. In the future, it
should be possible to give each guest it's own hard-disk controller, and
then the overhead would for disk-accesses would be eliminated - but any
hardware that is shared must be "shared safely", which means that
software needs to be involved in keeping track of what's going on in one
way or another - how and where can make a small difference, but it's
still overhead compared to the "non-virtual solution".
--
Mats
>
>
> My hardware is VT enable
>
> Thanks,
> Antoine N.
>
> _______________________________________________
> Xen-users mailing list
> Xen-users@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-users
>
>
>
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
|
Home