Re: H/W RAID 5 on slower disks versus no raid on faster HDDs

>>>>> "Nikolaus" == Nikolaus Dilger <nikolaus@dilger.cc> writes:

Nikolaus> SCSI320 in theory is twice as fast as SCSI160.  But the
Nikolaus> bottleneck will be the throughput of the individual disks.
Nikolaus> 15,000 rpm of course will be faster than 10,000 rpm.  More
Nikolaus> interesting then the rpm numbers itself are seek time and
Nikolaus> transfer rate.

More to the point, with current disks, SCSI160 needs 3 to 4 disks to
be saturated.  Don't buy 320 unless you have more than 4 disks.

Nikolaus> In a production environment I would always favor some kind
Nikolaus> of error protection. Either RAID 5 or RAID 1 (mirroring).  A
Nikolaus> hardware RAID controller is faster than software RAID.

I'm on a bit of a mission to stamp out this misconception.  In my
testing, all but the most expensive hardware raid controllers are
actually slower than FreeBSD's software RAID.  I've done my tests with
a variety of controllers with the same data load and the same disks.

As with any test, I have a theory: that the 2Ghz+ main processors of
modern machines so outstrip most raid controllers that it is faster to
perform the RAID on the main processor.  It is also lower latency
... and latency is what matters for advanced filesystems.

Nikolaus> For pure speed raw devices would be faster then file
Nikolaus> systems.  However, PostgeSQL currently does not support
Nikolaus> them.

This used to be true on machines with less processor power than disk
bandwidth.  It is likely no longer true.  To be more exact: yes,
filesystems have overhead, but the overhead is processor overhead
... of which (compared to disk bandwidth) you have lots.  OSs have
also become more efficient.

Dave.

--
============================================================================
|David Gilbert, Velocet Communications.       | Two things can only be     |
|Mail:       dgilbert@velocet.net             |  equal if and only if they |
|http://daveg.ca                              |   are precisely opposite.  |
=========================================================GLO================