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

Re: [Xen-users] questions regarding HVM and maximum block device size



> right now I run a bunch of PVM Xen guests, everything is fine. But on the
> horizon, there a potential need shows up, that I may have to run one or
> more HVM guests.
>
> Some time ago, I did some tests, and I observed the following on a host:
> I activated the AMD VT extension in the BIOS, because I wanted to test to
> setup a HVM machine. While this was activated, the PVM domU running on the
> same host, had a unusual slow NFS performance. After I was ready with the
> tests, I disabled the AMD VT in the BIOS again, and the NFS speed
> was "normal" again. The NFS speed with VT enabled was about 1/3rd slower
> than without. The dom0 and domU are 64Bit, SLES10SP1 systems.
> Is this normal what I've seen?

I don't think that's normal at all - it's certainly not the intended 
behaviour!  You're *just* running PV domains on the box, right?  The only 
difference is that you've enabled AMD-V in the bios?  That shouldn't make any 
difference at all, so it's most curious if there's a performance difference.

Have you also tried enquiring about this on SLES mailing lists / forums, in 
case it's a SLES-specific problem?

> If yes, I guess it's not recommended to run 
> PVM and HVM systems on the same dom0? Or if no, any idea, what I can do
> about it?

It should be fine to mix PV and HVM guests on the same system.  This is a 
pretty weird problem you're seeing though - I've no idea what would be 
causing it.  Are you sure that the bios setting is the only thing that 
changed?  Have you double checked your measurements here?  I don't mean to 
sound disbelieving, it's just a very very strange problem to see!

Assuming this is definitely reproducible, further enquiries are the way 
forward.  Asking on the SLES support channels makes sense.  Asking on 
xen-devel may also be worthwhile.

Check xm dmesg and /var/log/xen/xend.log for any differences in output between 
the two cases.  I don't know what I'd expect to see differ but it's worth a 
try.

> Further I'd like to know, whether a xm mem-set will work for HVM domU's?
> I guess, in case the OS supports it, then it will work?
> I've also read about paravirtual drivers for HVM guests, and I've seen a
> xen-balloon.ko for HVM Linux guests, but I want to run MS Windows, are
> there also such drivers available?

xm mem-set can work in principle for HVM domUs, yes.  AFAIK you won't be able 
to grow a domain beyond its initial allocation at this point in time but you 
should be able to shrink and grow it within those bounds.

You need an appropriate driver for the HVM OS though.  As you've noticed, 
there is a Linux driver available.  For Windows, you'll need to find some 
PV-on-HVM drivers for your platform.  I seem to recall Novell providing a 
driver pack for Windows on SLES - maybe you could look into that?  But 
there's also a free set of PV-on-HVM drivers, with the development being led 
by James Harper although I don't know if these have a balloon driver at this 
time...?  These are still in development, so they may not be recommended for 
use on a system containing important data or requiring high uptimes.  That 
said, I get the impression quite a few people are using them successfully 
having worked out any local problems.  Make sure to read through some mailing 
list archives on the drivers so you can learn of possible problems and 
actions to take to avoid them!

You may well want to experiment with PV-on-HVM anyhow to get better Windows IO 
performance.

> VMWare had, or still has, don't use it anymore since there is xen ;), a
> limit on the maximum size of a block device, at 2TB. So if I wanted to
> share a disk larger than 2TB, then the VMWare guest was/is only able to see
> the 2TB but not more. Does in Xen exists a similar limit on block device
> size?

I think there is a maximum block device size under Xen but I'm not sure what 
it is.  If you search the mailing list archives you may find some useful 
information on this.

Cheers,
Mark

-- 
Push Me Pull You - Distributed SCM tool (http://www.cl.cam.ac.uk/~maw48/pmpu/)

_______________________________________________
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®.