zfs discuss - Jul 2011 - ZIL, L2ARC, rpool -- partitions and mirrors oh my!

Putting together a server for a friend''s recording studio.
He''s planning
to do audio editing off the server, so low latency is a big deal. My plan
is to create a pool of two 8-drive RAIDZ2 vdevs and then accelerate
them... But how?

OS if going to be latest OpenIndiana. I have a pair of 40GB SSDs (Crucial)
with good write speeds and a pair of 64GB SSDs ( with good read speeds.
I''d like to mirror the root pool.

My initial thought was mirror the 40GB SSDs for the ZIL and partition the
two 64s; mirror two slices for the rpool and two slices for the L2ARC. If
there''s a smarter way to do it, suggestions gratefully accepted. My
current ZFS storage servers are all built around sustained reads/sustained
writes, so tuning the ZIL and L2ARC are still outside my experience.
-- 
Dave Pooser
Manager of Information Services
Alford Media Services, Inc.

On Thu, 7 Jul 2011, Dave Pooser wrote:
>
> My initial thought was mirror the 40GB SSDs for the ZIL and partition the
> two 64s; mirror two slices for the rpool and two slices for the L2ARC. If
> there''s a smarter way to do it, suggestions gratefully accepted.
My
> current ZFS storage servers are all built around sustained reads/sustained
> writes, so tuning the ZIL and L2ARC are still outside my experience.
Since you have been reading this forum for a long time, you should 
already know that the key issue for zfs will be when zfs writes a 
transaction group because it tends to freeze during the transaction. 
Otherwise zfs will perform well.  The key will be to tune the maximum 
TXG group size down to a size that your devices can handle 
expediently.  The tunable for this is zfs_write_limit_override.

For example put something like this in /etc/system:

* Set ZFS maximum TXG group size to 2684354560
set zfs:zfs_write_limit_override = 0xa0000000

You could also tune down zfs_vdev_max_pending:

* Set device I/O maximum concurrency
*
http://www.solarisinternals.com/wiki/index.php/ZFS_Evil_Tuning_Guide#Device_I.2FO_Queue_Size_.28I.2FO_Concurrency.29
set zfs:zfs_vdev_max_pending = 5

Bob
-- 
Bob Friesenhahn
bfriesen at simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/

> From: zfs-discuss-bounces at opensolaris.org [mailto:zfs-discuss-
> bounces at opensolaris.org] On Behalf Of Dave Pooser
> 
> low latency is a big deal. My plan
> is to create a pool of two 8-drive RAIDZ2 
You''ll get lower latency (and lower usable disk space) using mirrors
instead
of raidz.

Since you''re talking about a standalone file server, which is not doing
things like credit card transactions, you''ll get the highest
performance if
you disable the ZIL and forget about the dedicated log devices.  You can use
those devices as cache instead.