zfs discuss - Oct 2006 - zfs/raid configuration question for an application

Hi all,

I am going to have solaris 06/06, a database (postgresql), application server
(tomcat or sun app. server) on x86, 2GB with max. 3 scsi (hardware raid
possible) disks. There will be no e-mail or file serving.

The ratio of write/read operations to database is >=2 (and there is no other
significant disk io). Block size of postgresql is default (I think it is 8K).
Our application performance is bounded with database write (and then read)
operations, so db write/read performance is important.

I need redundancy in all over the system, not only for db data, but also os
itself. Capacity efficieny is not that important (I can use full mirroring).

Although I am not a sys.adm., I know some about raid levels and zfs, I want to
ask your comments on disk configuration.

Do you recommend mirroring or striping+some form of redundancy ? Specifically,
raid 1 or raid 5 ?
I am not going to use zfs as a root partition (since it is not possible in
solaris officially yet), so I need some ufs partition. So, do you recommend me
to use hardware raid or zfs ? If you recommend zfs, how can I configure it (I
mean which disk/slice as zfs, and which disk/slice as ufs) since I only have 3
disks ?

Thank you in advance.

Mete
 
 
This message posted from opensolaris.org

Mirroring will give you the best performance for small write operations.

If you can get by with two disks, I?d divide each of them into two slices, s0
and s1, say. Set up an SVM mirror between d0s0 and d1s0 and use that for your
root. Set up a ZFS mirror between d0s1 and d1s1 and use that for your data.

Any 3-disk configuration would be significantly more awkward and waste some
space.

For mirroring in your situation, I would tend to avoid hardware RAID. You won?t
be maxing out the controller rates so you won?t get much benefit.
 
 
This message posted from opensolaris.org

Quite helpful, thank you.

I think I should set the zfs mirror block size to 8K to match it with db, right
?
and do you think I should create another zfs mirror for transaction log of pgsql
? or is this only useful if I create zfs mirror on a different set of disks but
not slices ?

Mete
 
 
This message posted from opensolaris.org

Yes, set the block size to 8K, to avoid a read-modify-write cycle inside ZFS.

As you suggest, using a separate mirror for the transaction log will only be
useful if you''re on different disks -- otherwise you will be forcing
the disk head to move back and forth between slices each time you write.
 
 
This message posted from opensolaris.org

Hello Anton,

Thursday, October 12, 2006, 11:45:40 PM, you wrote:

ABR> Yes, set the block size to 8K, to avoid a read-modify-write cycle inside
ZFS.

Unfortunately it won''t help on 06/06 until patch is released to fix a
bug (not to read old block if it''s "overwritten"). However it
still is
wise to do it (set to 8k) ''coz when patch will be released it could
actually win some performance.

ps. it''s fixed in snv for some time

-- 
Best regards,
 Robert                            mailto:rmilkowski at task.gda.pl
                                       http://milek.blogspot.com

Since it is embedded into Solaris now, further improvements will be available by
patching and we dont have to migrate to another update, right ?

Two more things;
1) I know the benefits of ZFS, but I wonder if having ufs+zfs instead of only
ufs on single disk mirroring (I mean there are two disks, one is mirror)
configuration has real benefits (occam''s razor). Because, I usually see
people are using it with many disks.
2) Regarding to SW Raid & HW Raid solutions, the only thing I understand is
SW solutions may be implemented more efficiently due to talented OS engineers.
Is there any other benefit of SW Raid over HW Raid or viceversa for mirroring ?
HW Raid is quite transparent to OS, that is why I am questioning SW Raid.

Thank you, your comments are really helpful to me.

Mete
 
 
This message posted from opensolaris.org

Most ZFS improvements should be available through patches. Some may require
moving to a future update (for instance, ZFS booting, which may have other
implications throughout the system).

On most systems, you won?t see a lot of difference between hardware or software
mirroring.

The benefit of software mirroring is primarily that you don?t depend on a
controller. ZFS gives the additional benefit that not only a failed disk block,
but one which was written incorrectly, can be detected and recovered from the
alternate side of the mirror.

The benefit of hardware mirroring is twofold. First, the ?dirty map? can be
maintained in fast hardware (e.g. NVRAM), which can reduce the amount of time
that it takes to rebuild the mirror at startup and may slightly increase the
speed of random writes. (ZFS uses a different technique to maintain consistency
and does not need to rebuild its mirror after a crash, unlike SVM.) Second, you
only move the data once across the host bus and disk controller, instead of
twice, which on a heavily loaded system can increase your I/O throughput.
 
 
This message posted from opensolaris.org