[one-users] impossible to setup different transfer manager for system datastore on different hosts

Rolandas Naujikas rolandas.naujikas at mif.vu.lt
Tue Jun 26 12:49:13 PDT 2012


On 2012-06-26 20:39, Ruben S. Montero wrote:
> Hi,
>
> The rationale behind this is the following:
>
> The current datastore system allows you to setup a host that uses multiple
> datastores, each one with a different transfer driver. In this way, you can
> have FS datastores that is exported through a shared FS other FS datasores
> with SSH, and even one using an iSCSI server. With OpenNebula 3.4 you can
> use all of them at the same time in every single host (each host using
> tm_shared, tm_ssh, tm_iscsi depending on the image).
>
> In previous version you are restricted to a single TM for each host. This
> usually means for example that you are restricted to a single NFS export or
> iSCSI server. IMHO this is a clear gain on the storage subsystem.
>
> Now, the system datastore . It is used to create end-points in the target
> host, so the operations specific to the system datastore are just:
>
> context, mkimage, and mkswap: These by default create files for the ISO
> context CD-ROM or volatile disks
>
> mv: that mv's VM directories across hosts
>
> delete: to delete any temporal content created in the system datastore
>
> NOTE: clone, mvds, and ln operations are datastore specific, and we are not
> using the system ones.
>
> So I think that there is no regression. Note that the use of multiple
> system datastores will basically affect cold migrations (mv), which are not
> possible across hypervisors, and in general very limited across hosts with
> different configurations (i.e migrating a VM with a LVM as disk that need
> to be converted to a file in other host)

At least I saw that in some (commercial) cloud software, which really 
use VirtualBox/qemu-utils to convert different image formats for 
different virtualization platform.

Regards, Rolandas

P.S. Really I didn't test it too much, because I hate UI in browser made 
with Flash Player.

> However, I see situations where creating a context volume or a volatile
> volume in a LVM device or in a file depending on the host can be useful.
> So, probably a good trade-off would be setting up the system datastore per
> cluster instead of opennebula installation. What do you think?
>
> Thanks for your comments!
>
> BTW, Hope this helps you to tune the LVM2 drivers... Thanks also for that
> one :)
>
> Cheers
>
> Ruben
>
>
>
> On Tue, Jun 26, 2012 at 12:43 PM, Rolandas Naujikas <
> rolandas.naujikas at mif.vu.lt> wrote:
>
>> Hi,
>>
>> In opennebula 3.4.x there is not possible to setup different transfer
>> manager for system datastore on different hosts. That was possible in
>> opennebula 3.2.x and early. That looks like REGRESSION.
>>
>> That could be useful for opennebula with different visualization hosts
>> types (KVM, Xen, VMware) or different system datastore storage
>> configurations (filesystem + ssh/shared, filesystem + lvm2ssh/shared).
>>
>> Regards, Rolandas Naujikas
>> ______________________________**_________________
>> Users mailing list
>> Users at lists.opennebula.org
>> http://lists.opennebula.org/**listinfo.cgi/users-opennebula.**org<http://lists.opennebula.org/listinfo.cgi/users-opennebula.org>
>>
>
>
>





More information about the Users mailing list