Mailing List Archive: Diskspace to DomU

Diskspace to DomU

May 15, 2005, 6:01 PM

Post #1 of 3 (801 views)

Hi,

I am about to setup a new Xen server which will host a few DomU's, one of which will be a file server.
The file server will, naturally, have quite a lot of disk space and my question is:

If my file servrer domain uses a lot of disk space and the disks used are on two separate controllers, is it possible to dedicate those two PCI-devices to the domain and have that domain run LVM or EVMS to manage the disks itself instead of exporting a volume through Dom0?

And if it is, which is best? Does it yield better performance this way, or should I stick with exporting it through Dom0?

Oh, I have already used dedicated Ethernet NICs for a few domains so I now it is possible to assign PCI devices to privileged domains, so my first question is really if block devices is a "special case" which will not work.....

Any input is appreciated.

Best Regards
Jake S.

--

___________________________________________________________
Sign-up for Ads Free at Mail.com
http://mail01.mail.com/scripts/payment/adtracking.cgi?bannercode=adsfreejump01"]http://www.mail.com/?sr=signup

Re: Diskspace to DomU [ In reply to ]

mark.williamson at cl

May 15, 2005, 6:52 PM

Post #2 of 3 (796 views)

Permalink

> I am about to setup a new Xen server which will host a few DomU's, one
> of which will be a file server. The file server will, naturally, have
> quite a lot of disk space and my question is: If my file
> servrer domain uses a lot of disk space and the disks used are on two
> separate controllers, is it possible to dedicate those two PCI-devices to
> the domain and have that domain run LVM or EVMS to manage the disks
> itself instead of exporting a volume through Dom0?

Yes, it is. You can set up multiple backends for block devices just as you
can for network.

> And if it is,
> which is best? Does it yield better performance this way, or should I stick
> with exporting it through Dom0?

It may have slightly better performance (or at least lower CPU utilisation)
but depending on how heavily loaded the system is, it may not buy you
anything.

If the CPU on the machine is going to be particularly highly loaded and you're
going to be using high bandwidth disk IO, you may see some improvement. I'd
suggest that in the first instance you set up a test machine with all devices
driver by dom0 and see if that performs OK. If the performance is lower than
it should be, you might like to try assigning devices to the fileserver.

Cheers,
Mark

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

Re: Diskspace to DomU [ In reply to ]

tupshin at tupshin

May 16, 2005, 12:15 AM

Post #3 of 3 (787 views)

Permalink

Jake Stevenson wrote:

> Hi,
>
> I am about to setup a new Xen server which will host a few DomU's, one
> of which will be a file server.
> The file server will, naturally, have quite a lot of disk space and my
> question is:
>
> If my file servrer domain uses a lot of disk space and the disks used
> are on two separate controllers, is it possible to dedicate those two
> PCI-devices to the domain and have that domain run LVM or EVMS to
> manage the disks itself instead of exporting a volume through Dom0?
>
> And if it is, which is best? Does it yield better performance this
> way, or should I stick with exporting it through Dom0?
>
> Oh, I have already used dedicated Ethernet NICs for a few domains so I
> now it is possible to assign PCI devices to privileged domains, so my
> first question is really if block devices is a "special case"
> which will not work.....
>
> Any input is appreciated.
>
> Best Regards
> Jake S.
>
FWIW, my approach for this was to export the raw devices that make up
the lvm to my fileserver domU and exclude (raid devices md2, md3, and
md4 in my case) them from /etc/lvm/lvm.conf in the dom0:
filter = ["r|/dev/md2|", "r|/dev/md3|", "r|/dev/md4|" ]

my domU disk config then looks like this:
disk = [ 'phy:xen_vg/fileserver,hda1,w',
'phy:xen_swap_vg/fileserver,hda2,w',
'phy:md2,sda2,w',
'phy:md3,sda3,w',
'phy:md4,sda4,w',
]

I haven't done any comparative benchmarks of this approach vs. any
other, but I haven't noticed any aggregiously bad performance from my
fileserver domU either.

-Tupshin

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