On 09/11/2011 06:44 PM, Anthony Presley wrote:
> We've currently got PG 8.4.4 running on a whitebox hardware set up,
> with (2) 5410 Xeon's, and 16GB of RAM. It's also got (4) 7200RPM SATA
> drives, using the onboard IDE controller and ext3.
>
> A few weeks back, we purchased two refurb'd HP DL360's G5's, and were
> hoping to set them up with PG 9.0.2, running replicated. These
> machines have (2) 5410 Xeon's, 36GB of RAM, (6) 10k SAS drives, and
> are using the HP SA P400i with 512MB of BBWC. PG is running on an
> ext4 (noatime) partition, and they drives configured as RAID 1+0
> (seems with this controller, I cannot do JBOD). .
> To start with, I've set the "relevant" parameters in postgresql.conf
> the same on the new config as the old:
>
> fsync = on
> synchronous_commit = off
The main thing that a hardware RAID controller improves on is being able
to write synchronous commits much faster than you can do without one.
If you've turned that off, you've essentially neutralized its primary
value. In every other respect, software RAID is faster: the CPUs in
your server are much faster than the IO processor on the card, and Linux
has a lot more memory for caching than it does too. Turning off sync
commit may be fine for loading, but you'll be facing data loss at every
server interruption if you roll things out like that. It's not
realistic production performance for most places running like that.
A lot of your test results seem like they may be using different levels
of write reliability, which makes things less fair than they should be
too--in favor of the cheap IDE drives normally. Check out
http://wiki.postgresql.org/wiki/Reliable_Writes for more information
about that topic.
--
Greg Smith 2ndQuadrant US greg@2ndQuadrant.com Baltimore, MD
PostgreSQL Training, Services, and 24x7 Support www.2ndQuadrant.us
