zfs discuss - Oct 2006 - Migrating vdevs to new pools

Howdy,

We use VxVM quite a bit at my place of employment, and are extremely 
interested in moving to ZFS to reduce complexity and costs. One useful 
feature that is in VxVM that doesn''t seem to be in ZFS is the ability
to
migrate vdevs between pools. This is extremely useful when you want to 
synchronize a set of temporary disks with a set of master disks, and 
migrate the temporary disks to an alternate server once the
synchronization completes (we use this to perform host free backups). Are 
there any plans to introduce a similar capability in ZFS? I read through 
the zpool manual page and the ZFS documentation on opensolaris.org, but 
was unable to find a specific set of options to accomplish this. If I 
missed something, please let me know. I will glady wander off to read.

Thanks for any insight,
- Ryan
--
UNIX Administrator
http://prefetch.net

Matty wrote:
> We use VxVM quite a bit at my place of employment, and are extremely 
> interested in moving to ZFS to reduce complexity and costs. One useful 
> feature that is in VxVM that doesn''t seem to be in ZFS is the
ability to
> migrate vdevs between pools. 
Could you be more specific?  Are you moving disks between disk groups?
Or, are you moving disk groups between machines?  I suspect that there is
a terminology barrier here which could send us down a rathole...
>                              This is extremely useful when you want to 
> synchronize a set of temporary disks with a set of master disks, and 
> migrate the temporary disks to an alternate server once the
> synchronization completes (we use this to perform host free backups). 
> Are there any plans to introduce a similar capability in ZFS? I read 
> through the zpool manual page and the ZFS documentation on 
> opensolaris.org, but was unable to find a specific set of options to 
> accomplish this. If I missed something, please let me know. I will glady 
> wander off to read.
zfs send/receive is likely what you want.  The main difference between
copying mirrors (ala VxVM) and zfs send/receive is that in the VxVM case
you copy all of the disk blocks, whereas in ZFS you only copy the data.

My first impression is that you want a zpool which can move between
systems (similar to a VxVM disk group), and would use zfs send/receive to
"copy" the data between zpools (locally or remotely, as an option).
I don''t see any advantage, but a few disadvantages, to copy by
silvering
mirrors.
  -- richard

> > We use VxVM quite a bit at my place of employment, and are extremely 
> > interested in moving to ZFS to reduce complexity and costs. One useful
> > feature that is in VxVM that doesn''t seem to be in ZFS is the
ability to
> > migrate vdevs between pools. 
> 
> Could you be more specific?  Are you moving disks between disk groups?
> Or, are you moving disk groups between machines?  I suspect that there is
> a terminology barrier here which could send us down a rathole...
Besides other things, I think he''s talking about the ability VxVM has
to

a) Take the redundant data from an existing mirror and disconnect it so
   that it is visible as an independent volume.

   I''m not sure exactly how ZFS manages mirrors, so I don''t
know how
   hard this would be to achieve.  Are the same issues surrounding disk
   evacuation in play here?  I think it''s different because you can
   remove and attach mirrors.

b) Assuming the new mirror is on a set of disks with no other data,
   split the volumes (the original and the new) into separate
   diskgroups (allowing the disks to be moved to another host with the
   data intact).

   I think this is difficult.  If possible, I would guess it would have
   to be done as a single step with a).  If the split volume were ever
   brought up as a live member of the original pool, there would be no
   way to separate the data that lived on the members of the pool.
> zfs send/receive is likely what you want.  The main difference between
> copying mirrors (ala VxVM) and zfs send/receive is that in the VxVM case
> you copy all of the disk blocks, whereas in ZFS you only copy the data.
It depends on how it''s set up.  VxVM can copy a subset of blocks to
recreate mirror consistency if DSO logs are in use.
> My first impression is that you want a zpool which can move between
> systems (similar to a VxVM disk group), and would use zfs send/receive to
> "copy" the data between zpools (locally or remotely, as an
option).
> I don''t see any advantage, but a few disadvantages, to copy by
silvering
> mirrors.
To me, the advantage appears to only be the number of disks required for
the copy.  ZFS would require either 3 (mirror plus destination), or 2
and time to copy the entire dataset.  VxVM can use 2 with no time to
copy other than resilvering after the mirror is re-established.

I doubt that''s enough of a feature to want to spend as much time as I
assume would be required to fit it into ZFS as it exists.



-- 
Darren Dunham                                           ddunham at taos.com
Senior Technical Consultant         TAOS            http://www.taos.com/
Got some Dr Pepper?                           San Francisco, CA bay area
         < This line left intentionally blank to confuse you. >