Our server team has set up an EMC Clarion SAN with 3 sets of RAID 5 arrays,
several RAID 1 arrays and other drives/arrays for snapshots. The first
server we are installing has the data going to an 80 GB slice an an 8 disk
RAID 5 array and the logs going to a 20 GB slice on a 5 disk RAID 5 array.
Eventually, several other database servers will be added to this SAN.

The first server will hold several small databases and one or more large
databases, all totaling 20 - 50 GB. Some of the large databases will be
replicates from other servers. I assume the replication will be
transactional with non updateable subscriptions. Some of the databases will
also replicate to other servers. I assume the replication for these will be
transactional with updateable subscriptions. All databases will be serving
up content for web applications.

Does anyone have experience using RAID 5 on an EMC Clarion for the log
files?

I do not know how much data will be replicated and I don't have any stats on
the servers the app was developed on. Would anyone hazard a guess as to the
MB/hour of data that may overwhelm this RAID 5 data volume?

Any educated opionions are welcome, especially if you can back it up with
some examples etc.

I avoid RAID 5 at all cost. See http://www.baarf.com/ if you are bored and
wanna read some experiences, horror stories, sillies, rants, etc..

That said, the nature of your application plays a huge role in the decision
on what RAID configuration. Stripping with parity is probably "ok" if you're
doing mostly reads and have solid backup and maintenance processes in place
(eg. swap out disks before MTBF). However, if all all possible, put your
data on a RAID 1+0. Yes, quite a bit of extra cost involved but if you're
paranoid about simultaneous disk failures and/or have high performance
requirements, definitely worth considering. I've done this stuff for 15
years and have only seen 1 CPU failure ever. I have seen a good number of
multiple drive failures either simultaneously or very close to each other.
There were 2 occurrence of a second drive failing shortly after a
replacement drive was stuck into the array and rebuilding had only just
begun. Good thing we had backups.

Transaction logs are perfectly happy sitting in its own disk or mirrored
disk for availability. I would not recommend RAID 5 for your logs. It's just
an unecessary performance hit (tlogs activities are sequential) plus it'll
use up more physical drives than a mirror anyway.

One thing you want to be careful with is what the SAN administrators are
really giving you when they assign you a LUN. There's a lot of black magic
inside a SAN box and I have met many SAN vendor reps that insist it doesn't
matter if we're getting a "slice" of an array or our own array because the
big caches. These guys obviously have not had to work with high transaction
environments with multiple workloads on the same SAN.

Btw, there are free tools available to stress your IO. See below for one
example from Microsoft.
http://www.microsoft.com/downloads/d...DisplayLang=en



joe.

"Dean" <deann (AT) dtn (DOT) com> wrote

Push back. We just went through the same issue with our local EMC technical
sales guy when we bought our CX-700.

It seems to me he tried to talk our SAN guy into going with a bunch of 4
drive RAID 5 arrays because he could show better MB/$ costs. His hope was
that they could make up for the RAID 5 overhead with their cache. When I
pushed back, he brought in their regional SQL Server specialists (that's how
they were positioned anyway). When we explained our throughput requirements
(the same ones we explained to our local tech rep), they both said, "You
need RAID 10".

"Dean" <deann (AT) dtn (DOT) com> wrote