Raid + zfs performace.

Jeremy Chadwick freebsd at jdc.parodius.com
Sat Oct 30 20:11:49 UTC 2010


On Sat, Oct 30, 2010 at 09:13:41PM +0200, Peter Ankerstål wrote:
> > 
> > 
> >> Now you presented me with a third option. So you think I should skip to create
> >> a new hardware-raid mirror and instead use two single drives and add these as
> >> a mirror to the existing pool?
> > 
> > If you're going to keep the hardware raid, then setting up a new
> > hardware raid of two drives, and then striping da1 with da0 via zfs is
> > a viable option. It's just another spin on the RAID 10 idea.
> > 
> Sorry to ask again but I'm still not sure what you think is the best solution when 
> comparing adding the two new drives as a zfs mirror like:
> pool
> 	da0
> 	mirror
> 		da1
> 		da2
> 
> or making a hardware mirror da1 and adding that one 
> 
> pool 
> 	da0
> 	da1

The answer is "it depends", and I can't authoritatively act as your
system administrator since I don't have any familiarity with what it is
your systems are doing and so on.  That's your job.  :-)  You'd need to
disclose exactly:

- What hardware RAID controller you're using and all of its
  capabilities, including if it has cache and a BBU,
- Full details of the workload on the machine and what the majority of
  I/O consists of,
- What exact OS you're running (uname -a please) and how much physical
  RAM the system has.

If you really want to answer your own question, I would recommend at
least performing benchmarks (bonnie++ might suffice) with both setups.
And don't forget that if you use ZFS you'll need to perform some
minor loader.conf tuning, and expect to adjust values depending on
workload/environment.

> And by the way. you guys seem zfs-shifty.

Language barrier detected!  :-)  "ZFS-shifty" could mean either "you're
ZFS advocates (fans of ZFS and recommend using it over anything else)",
or "you're timid when it comes to/afraid of ZFS".  I think you meant the
first one, but I'm not certain.

If so: believe it or not, I'm not much of a FreeBSD ZFS advocate. There
are issues that keep appearing on the mailing lists (-stable and -fs),
and each incident has to be handled individually.  There are definitely
stability issues (we just experienced one ourselves which was major[1];
it's been fixed in RELENG_8 since mid-October) which are still getting
hammered out.

My logic is this, and this is just one man's opinion:

- If you need absolute stability, don't have time or the desire to
  tinker with new technology (or have 100% mission-critical services
  in use), stick with using UFS + softupdates.
- If filesystem administrative simplicity is needed over everything
  else, ZFS is an excellent choice.
- If you want ZFS and need absolute rock-solid performance, stability,
  and It Should Just Work(tm), run Solaris 10 or OpenSolaris.
- If you're going to use ZFS on FreeBSD, you need to run RELENG_8, and
  should almost certainly be running amd64, and have at least 4GB RAM.

> Do you have any ideas about my other problem i posted to the list?
> (http://lists.freebsd.org/pipermail/freebsd-fs/2010-October/009922.html)

Nope, I don't.  I don't use ZFS send/recv nor snapshot capability.  I do
keep seeing problems reported with both of these on the lists, but
again, they have to be handled on a per-case basis.

[1]: http://lists.freebsd.org/pipermail/freebsd-fs/2010-October/thread.html#9687
     ("Locked up processes after upgrade to ZFS v15")

-- 
| Jeremy Chadwick                                   jdc at parodius.com |
| Parodius Networking                       http://www.parodius.com/ |
| UNIX Systems Administrator                  Mountain View, CA, USA |
| Making life hard for others since 1977.              PGP: 4BD6C0CB |



More information about the freebsd-fs mailing list