zfs discuss - Nov 2005 - Database on ZFS

We currently use raw devices (striped across 5 disks) for our database volumes
and I was wondering how ZFS can be used i.e. do I need to go to cooked files for
our database to reap the benefits of ZFS?

-- prasad
This message posted from opensolaris.org

On Thu, 2005-11-17 at 03:01, prasad wrote:
> We currently use raw devices (striped across 5 disks) for our database
volumes and I was wondering how ZFS can be used i.e. do I need to go to cooked
files for our database to reap the benefits of ZFS?
This is one of the use cases for zvols.

See the zfs(1m) man page for how to create them, example number 12
but instead of using swap -a configure your database to that device.

-- 
Darren J Moffat

> This is one of the use cases for zvols.
Both will work -- zvols or zfs filesystems.

Either way, the key to good database performance is to match the zvol''s
blocksize (or the filesystem''s recordsize) to the database''s
record size.
That way your database writes will be pure writes, not read-modify-write.

For a 100G database using 8k blocks, create your zvol like this:

# zfs create -V 100g -b 8k pool/volume

If you need to change the size, just set the ''volsize''
property:

# zfs set volsize=200g pool/volume

Alternatively, if you want to use a filesystem and just store the database
in a file, that''ll work too:

# zfs create pool/fs
# zfs set recordsize=8k pool/fs

and you''re good to go.

Jeff
This message posted from opensolaris.org