Does RAID5 performance increase with additional drives?

Or, put another way, will a hardware RAID5 system (such as running off
of an Adaptec FastTrak S150-SX4 SATA controller) perform better with
four 120-GB drives than it would be with three 160-GB drives, assuming
comparable drive performance specs?

It seems to me performance should improve with more devices (the
stripes are more distributed) but I can't seem to corroborate my
assumption with any of the materials I've reviewed...

Thanks!

On 16 Mar 2004 22:16:40 -0800, IneverReadAnythingSentToMe RemoveThis @hotmail.com
(David Filmer) wrote:

 >Or, put another way, will a hardware RAID5 system (such as running off
 >of an Adaptec FastTrak S150-SX4 SATA controller) perform better with
 >four 120-GB drives than it would be with three 160-GB drives, assuming
 >comparable drive performance specs?
 >
 >It seems to me performance should improve with more devices (the
 >stripes are more distributed) but I can't seem to corroborate my
 >assumption with any of the materials I've reviewed...

Yes. Normally the performance of Raid5 will increase when you have
more disks.

But if that it also noticable in this case remains to be seen.

When you have more disks the large file can be spread over more disks
and the transfer rate increases. This would happen if you use four
instead of three drives.

But
- the PCI interface might well be the bottlneck in this case. Most
people will have the S150 SX4 in a 32/33 pci slot and with three disks
you can already reach the limit of that slot. But that also depends if
your the file is located on the faster part of the drive and how
fragmented it is.
- Transfer rate isn't all that important for disk performance anyway,
so whether you will notice even the maximum possible increase is
doubtful

Another far more important reason why Raid5 is faster when you have
more disks is because the "effective" access time decreases.
I'll try to give a very simple explanation using Raid0:
suppose you have 4 disks in a Raid0 array. And you are reading lots of
small files that are each spread over 2 disks.
When you read the first file, only 2 drives in the array are busy.
Instead of letting the other 2 drives do nothing you can let them
retrieve another file that is only located on those disks.
This way you can read 2 files in the time it normall takes to read 1
file. So you have effectively halved your access time.

Of course it is far easier to do this when you have lots of drives in
a array. when you have a 400GB array with 4 disks you have much less
chance of being able to make use of this, then when you have a 400GB
array with 40 disks.

In your situation where you have a small number of harddisks this
would only be possible for very small files. I Assume you have set the
stripe size to the largest possible value. So you would only benefit
if you read lots of files about the size of 64kB.

So to make a long story short: I think you will see a small increase
in benchmarks, but not in normal use.

Marc<!-- ~MESSAGE_AFTER~ -->