[OpenIndiana-discuss] ZFS storage best practices

[Dan Swartzendruber]

Gabriele Bulfon wrote:
> Hello, I have some questions about best strategies when using OI as a ZFS storage.
> My general idea, is that servers should boot from zfs volumes shared via iscsi.
>   
One caveat: rather than a zvol, I would use a thin-provisioned file to 
back the iSCSI LUN.  People (including me) have experienced inferior 
performance using the zvol...
> In the same general idea, data shoud reside on a native zfs file system of the storage, so that
> I can run snapshots from the storage, and find back my backup files easily inside the .zfs filesystem.
> In this scenario, data files can be easily recovered, while system files will need zfs clones and temporary sharing, to be recovered.
> This sounds ok, as data files are in the hands of users, and often need to be restored, while system files usually needs to be restored because of serious problems, by IT staff, and hopefully seldom.
> For windows data, one should use CIFS to share the storage pool to windows users directly, on a net trunk connected to the corporate LAN.
> For windows databases and similar files that cannot reside on a shared filesystem, the db can be hosted on an iscsi volume, then backup files can be sent to a native zfs file system, again to be easily restored.
> Then comes the decision for different servers' data.
> If I have a linux or another solaris server using the storage (a mail server, for example), using iscsi volumes would not let me take advantage of the quick restore of a single file, expecially for linux.
> So, my idea would be to use a native zfs pool, shared via nfs.
> But...someone is telling me that this will be much slower.....
>   
It all depends, I guess.  Some people get better performance with iSCSI, 
others with NFS.  The big advantage of going NFS is that the OI server 
can back up user data by snapshotting the zfs filesystem on the OI 
server (this is not the only kind of backup you want, of course...)
> In case this is true, linux will have only this chance, to have the easy of single file recovery, running slower.
> In case this is true, another solaris server may attach an iscsi volume, and crate its own zfs filesystem on it, and run its own snapshots on it, and have easy recovery.
> But....is this a good choice?
> I understand I'm making a zfs on a zfs volume, and I don't know if this may sound a good decision...
> Also, if the storage zfs is safe beacuse it's running on mirrored/raidz disks, then what about the server's zfs?
> I read everywhere that zfs always have to run on any kind of replication (mirror or zfs) to be safe.
> So what about the server's zfs? I can't think of doubling again the volumes to have another mirror...isn't it? How safe would be this?
> Hope someone would like to share these thoughts...
>   
I'm not sure what you mean above (about duplicating the duplicated data?)