I''m trying to guess what is the best practice in this scenario: - let''s say I have a zfs based storage (let''s say nexenta) that has it zfs pools and volumes shared as iScsi raw devices - let''s say I have another server running xvm or virtualbox connected to the storage - let''s say one of the virtual guests is OpenSolaris My question is: - is it correct to mount the iScsi device as base disks for the VM and then create zpools/volumes in it, considering that behind it there is already another zfs? - what alternatives do I have? - in case it''s correct to have the VM zfs over the storage zfs, where should I manage snapshots? on the VM or on the storage? Thanks for any idea Gabriele. -- This message posted from opensolaris.org
On Sat, May 1, 2010 at 7:08 AM, Gabriele Bulfon <gbulfon at sonicle.com> wrote:> My question is: > - is it correct to mount the iScsi device as base disks for the VM and then create zpools/volumes in it, considering that behind it there is already another zfs?Yes, that will work fine. In fact, zfs checksums will help protect from over the wire errors. You can enable redundancy at either or both levels, depending on performance requirements, available space and your level of paranoia. Using mirroring or raidz in your VM will use more bandwidth to your iscsi server.> - in case it''s correct to have the VM zfs over the storage zfs, where should I manage snapshots? on the VM or on the storage?It''s up to what you plan on doing with the VM. I''d probably do both, depending on the changes that I plan on making. For instance, use time slider / zfs-auto-snapshot on the VM, but also snapshot the zvol on the backing store before making any big configuration changes. -B -- Brandon High : bhigh at freaks.com
Thanks for your suggestions :) Another thing comes to my mind (expecially after a past bad experience with a buggy storage non-zfs backend). Usually (correct me if I''m wrong) the storage will be having redundancy on its zfs volumes (be it mirror or raidz). Once the redundant volume is exposed as a single iScsi volume, the virtual solaris will create his own zfs filesystem on it (unaware of its redundant backend). One of the "best practices" I''ve read specifically tells that single resource pools are unsafe. In this case, the backend knows it''s actually redundant, but the virtual os does not, and actually have just a single resource mounted as its zfs disk. Is this situation safe? Should I expose two iScsi volumes and let the virtual os again use a redundant zpool on them? This would obviously double again the disk requirements... Thanks again for any idea. Gabriele. -- This message posted from opensolaris.org