Software RAID issues (was Re: Suggestions solicited, server bring up)

[Bill McGonigle]

> On 23-Nov-2009, Alan Johnson <alan at datdec.com> sent:

> Nope. As I understand it, when you do an iSCSI export of a ZFS
> pool, you're getting a block device with the advantages of the
> ZFS storage mechanism without any particular filesystem on it.
>
> I could be wrong, of course. I haven't played with that part
> of ZFS yet.

Quite right.  You get the block management of ZFS but not the nice
filesystem semantics.  You're going to commit to a chunk of storage, so
the clients' filesystem sees something of the right size.  If you can get
away with NFS, you also get ZFS's dynamic filesystems, which are not
size-committed on creation (so you can share free space among
filesystems).

On linux, to use ZFS over iSCSI you used the iscsid software which
provides you with a /dev/sdc, e.g..  You treat that just like any other
block device (though you should mount it through /sys with the iscsid
semantics so that when [random thing] happens your drives don't get out of
order.

Where I'm using Xen, I run iscsid in Dom0 and provide the iSCSI target to
Xen as a phy:.  Supposedly Xen has better iSCSI support built in, but
maybe a newer version than what's in RHEL is needed for that to work.
Note: Dom0 can become CPU starved if you don't tell the scheduler to
prevent that.

-Bill