Hardware vs software raid

[Preston Hagar]

On Thu, Oct 20, 2011 at 1:04 PM, Nick Fox <nickj.fox at gmail.com> wrote:
> I can tell you from experience running an open bravo server you will want
> all the hardware performance you can get. Go hardware raid.
>

Sometimes software raid can be faster.  Since Open Bravo uses a
postgresql backend, I found a Postgresql benchmark:

http://wiki.postgresql.org/wiki/HP_ProLiant_DL380_G5_Tuning_Guide

If you look at the data, in many cases, software raid beat out
hardware RAID.  Often hardware wins out in writes, where software can
often win in reads, so it really depends on if you are doing a lot of
writes or not.

Also, if speed is that vital, instead of spending the extra several
hundred for a hardware RAID card, get more drives and go RAID 10.  You
get the redundancy of RAID 1 with speed improvements of RAID 0.
Again, it will really depend on the server, but with the speed and
drop in cost of RAM and CPUs (espeically multi-core CPUs), I have seen
many setups where software RAID can beat out a lot of hardware RAID
cards.

I guess if a minimal performance is vital, then setting both up and
doing benchmarks is probably the only way to be sure you are getting
the most performance you can out of your hardware.

Really though, with Open Bravo, what makes it slow is Tomcat and all
that JAVA, which is generally more RAM and CPU intensive and not I/O
intensive.  Most of the I/O will be Postgresql, which can be tuned to
run great even with a little bit slower drives.

I am sure everyone has their own experiences, the only thing I would
stress to the OP is to test out failure scenarios, because that will
be the time you want your system to really be good to work with.  I
have had too many headaches with hardware RAID failues (with PERC
cards, Areaca cards, 3ware cards, and Adaptec cards) to really go
through the stress anymore.  If I have one of those cards, I will
typically just use it as pass-through/JBOD and then do software RAID,
although that can even present a problem.  Again though, most of my
setups are small business, with a handful of servers, not large
setups, so I could see it being different for a large business.