Xen users - Nov 2006 - Newbe wants to move block device virtual machine

Hi list,

We are going to use Xen for our virtualization effort. Now I heard from some 
friends who are experienced Xen users that performance of the block device 
Xen machines is MUCH better than the performance of "file based"
virtual
machines. (is that true?)

Now if I use a block device as a basis for a virtual machine (on LVM), than it 
is not easy to move such a vm to another machine and make it accessible by 
iScsi. (is that true?)

At least I haven got the foggiest idea how to do that.

Anyone any ideas???

Cheers,
Dick Kniep

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

On Tue, Nov 21, 2006 at 06:15:30PM +0100, Dick Kniep
wrote:
> Hi list,
> 
> We are going to use Xen for our virtualization effort. Now I heard from
some
> friends who are experienced Xen users that performance of the block device 
> Xen machines is MUCH better than the performance of "file based"
virtual
> machines. (is that true?)
If you use blktap rather than loop devices for your file based virtual machines
there should be minimal difference, because blktap uses async/direct IO to
bypass
the FS layer caching. You can switch to blktap by using
''tap:raw:'' instead of
''file:'' when configuring the disks.  The other benefit of
using blktap for
file based disks is that it gives you much better data integrity upon crashes.
> Now if I use a block device as a basis for a virtual machine (on LVM), than
it
> is not easy to move such a vm to another machine and make it accessible by 
> iScsi. (is that true?)
If you want to migrate VMs, then you need to use a network aware block device
such as iSCSI, or GNDB, or a shared SAN with Cluster-LVM. Of course same
applies for file  backed disks - you need to make sure the file is shared
across all hosts to which you migrate. So either way you need shared storage
at some point or another.

Regards,
Dan.
-- 
|=- Red Hat, Engineering, Emerging Technologies, Boston.  +1 978 392 2496 -=|
|=-           Perl modules: http://search.cpan.org/~danberr/              -=|
|=-               Projects: http://freshmeat.net/~danielpb/               -=|
|=-  GnuPG: 7D3B9505   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505  -=| 

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

Hi Daniel,

Thanks for your answer, however there are some snags...

I am using Suse Linux Enterprise server version 10.0 which supports Xen 3.0.2 
where blktap is not supported, so I guess I am stuck with block devices.

Furthermore I will try to make clear what I want with the migration part:
We are an ASP company that delivers an ERP system to our users by means of 
NoMachine (or GoGlobal). Sales are picking up and we have to upgrade. Now we 
want to make several virtual machines that are all connected to a SAN. On the 
SAN which is used with iScsi, the virtual machines are stored. However....
The SAN is located in our hosting centre, while I am at the office to create 
the virtual machines. So what I want is to define a virtual machine here 
locally on a server, and store that virtual machine on the SAN later when I 
place the server in the rack. Is that possible?

Op dinsdag 21 november 2006 18:21, schreef Daniel P.
Berrange:
> On Tue, Nov 21, 2006 at 06:15:30PM +0100, Dick Kniep wrote:
> > Hi list,
> >
> > We are going to use Xen for our virtualization effort. Now I heard
from
> > some friends who are experienced Xen users that performance of the
block
> > device Xen machines is MUCH better than the performance of "file
based"
> > virtual machines. (is that true?)
>
> If you use blktap rather than loop devices for your file based virtual
> machines there should be minimal difference, because blktap uses
> async/direct IO to bypass the FS layer caching. You can switch to blktap by
> using ''tap:raw:'' instead of ''file:''
when configuring the disks.  The other
> benefit of using blktap for file based disks is that it gives you much
> better data integrity upon crashes.
>
> > Now if I use a block device as a basis for a virtual machine (on LVM),
> > than it is not easy to move such a vm to another machine and make it
> > accessible by iScsi. (is that true?)
>
> If you want to migrate VMs, then you need to use a network aware block
> device such as iSCSI, or GNDB, or a shared SAN with Cluster-LVM. Of course
> same applies for file  backed disks - you need to make sure the file is
> shared across all hosts to which you migrate. So either way you need shared
> storage at some point or another.
>
> Regards,
> Dan.
-- 
Met vriendelijke groet,

Dick Kniep
Lindix bv
De Specerij 30
1313 NE  Almere
tel. 036-521 5580
fax. 036-844 1825



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

Dick Kniep wrote:
> The SAN is located in our hosting centre, while I am at the office to
create
> the virtual machines. So what I want is to define a virtual machine here 
> locally on a server, and store that virtual machine on the SAN later when I
> place the server in the rack. Is that possible?
>   
I guess the easiest way would be to create the machines on filesystems,
place the server in rack, create LVMs there and restore the domU via tar
there.
You will just have to slightly modify the vm config file there (or use a
template).

Jan

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