[Dovecot] Maildir over NFS

Stan Hoeppner stan at hardwarefreak.com
Sun Aug 8 07:50:28 EEST 2010


CJ Keist put forth on 8/7/2010 10:17 AM:
> All,
>     Thanks for all the information.  I think I'm leaning towards locally
> attached fiber disk array.  Couple of advantages I see, one it will be
> faster than NFS, second it will allow us to separate user home directory
> disk quotas and email disk quotas. Something we have been wanting to do
> for awhile.

If you're going to do locally attached storage, why spend the substantial
additional treasure required for a fiber channel array and HBA solution?
You're looking at a minimum of $10k-$20k USD for a 'low end' FC array
solution.  Don't get me wrong, I'm a huge fan of FC SANs, but only when it
makes sense.  And it only makes sense if you have multiple hosts and you're
slicing capacity (and performance) to each host.  So at least get an entry
level Qlogic FC switch so you can attach others hosts in the future, or even
right away, once you realize what you can do with this technology.  If you're
set on the FC path, I recommend these components, all of which I've used and
are fantastic products when great performance and support:

http://www.qlogic.com/Products/SANandDataNetworking/FibreChannelSwitches/Pages/QLogic3800.aspx
http://www.sandirect.com/product_info.php?products_id=1366

http://www.qlogic.com/Products/SANandDataNetworking/FibreChannelAdapters/Pages/QLE2460.aspx
http://www.sandirect.com/product_info.php?cPath=257_260_268&products_id=291

http://www.nexsan.com/sataboy.php
http://www.sandirect.com/product_info.php?cPath=171_208_363&products_id=1434

Configure the first 12 of the 14 drives in the Nexsan as a RAID 1+0 array, the
last two drives as hot spares.  This will give you 6TB of usable array space
sliceable to hosts as you see fit, 600 MB/s of sequential read throughput,
~1000 random seeks/sec to disk and 35k/s to cache, ultra fast rebuilds after
drive failure (10x faster than a RAID 5 or 6 rebuild).  RAID 1+0 does not
suffer the mandatory RAID 5/6 read-modify-write cycle and thus the write
throughput of RAID 1+0 has a 4:1 advantage over RAID 5 and an 8:1 advantage
over RAID 6, if my math is correct.

The only downside of RAID 1+0 compared to RAID 5/6 is usable space after
redundancy overhead.  With our Nexsan unit above, RAID 5 with two hot spares
will give us 11TB of usable space and RAID 6 will give us 10TB.  Most people
avoid RAID 5 these days because of the "write hole" silent data corruption
issue, and go with RAID 6 instead, because they want to maximize their usable
array space.

One of the nice things about the Nexsan is that you can mix & match RAID
levels within the same chassis.  Let's say you're wanting to consolidate the
Postfix queues, INBOX and user maildir files, _and_ user home directories onto
your new Nexsan array.  You want faster performance for files that are often
changing but you don't need as much total storage for these.  You want more
space for user home dirs but you don't need the fastest access times.

In this case you can create a RAID 1+0 array of the first 6 disks in the
chassis giving you 3TB of fast usable space with highest redundancy for the
mail queue, user INBOX and maildir files.  Take the next 7 disks and create a
RAID 5 array yielding 6TB of usable space, double that of the "fast" array. We
now have one disk left for a hot space, and this is fine as long as you have
another spare disk or 2 on the shelf.  Speaking of spares, with one 14 drive
RAID 1+0, you could actually gain 1TB of usable storage by using no hot spares
and keeping spares on the shelf.  The likelihood of a double drive failure is
rare, and with RAID 1+0 is would be extremely rare for two failures to occur
in the same mirror pair.

-- 
Stan


More information about the dovecot mailing list