RAID controller question...

[Stanley, Jeremy]

So far it appears that, for me, ICP Vortex is winning over DPT at a raw
price level.  Admittedly this is not a very broad test since I'm only
looking at internal PCI-SCSI controller cards (because my array is not
hot pluggable, making the array-mount SCSI-SCSI impossible, and the
rack-mount external boxes are a lot more expensive) and then only those
that support low-voltage differential SCSI-3 drives (since that's what
I've got).

The ICP Vortex I'm looking at is the GDT6518RD, which has been priced to
me for $1245 by RAID Inc.

The DPT I'm considering is their PM3334UDW which they list for $1660,
but I might find it cheaper if I can locate a reseller...

The Mylex controllers I looked at were of questionable compatibility for
my array, and their answer-man is on vacation for a week, so they're
conspicuously absent from my list for now.
-------------------------*-------------------------
Jeremy J. Stanley
Information Systems Administrator
Trend Community Mental Health Services


> ----------
> From: 	Christian Worley[SMTP:cworley@altatech.com]
> Sent: 	Tuesday, September 08, 1998 3:11 PM
> To: 	Michael Pearl
> Cc: 	'Beowulf Cluster Mailing List'; 'Extreme Linux Mailing List'
> Subject: 	Re: RAID controller question...
> 
> Michael Pearl wrote:
> > 
> > Have you looked at DPT's controllers?  I don't have performance
> numbers,
> > but I am currently running RAID-5 using a 2144UWR with RedHat 5.0.
> > 
> 
> Michael,
> 
> Your system is in use, so performance testing may not be possible...
> 
> But, if you can destructively test your system, try (assuming your
> raid is /dev/sda):
> 
> 	time dd if=/dev/zero of=/dev/sda count=65536 bs=65536
> 
> That will destroy any file system on /dev/sda, so don't do it on
> anything but a new, destroyable, system (i.e., where you haven't run
> "mke2fs" yet).
> 
> A similar test via a file system, assuming it's mounted on /dev/hd,
> would be
> 
> 	time dd if=/dev/zero of=/mnt/hd/foo count=65536 bs=65536
> 	rm /mnt/hd/foo
> 
> This would not be a destructive write test, but would include the
> overhead of the file system and create a 4GB file.  Make sure to
> delete the file: if you run this test twice without deleting the file
> in-between, then you'll also measure the time for the file system to
> delete such a file, which is not insignificant.
> 
> A read test, that won't destroy the system and won't be worst case
> (because it won't show the overhead of calculating RAID-5 parity
> stripes) would be:
> 
> 	time dd if=/dev/sda of=/dev/null count=65536 bs=65536
> 
> But that figure would be interesting too.
> 
> I test performance using the raw device since this uses only their
> driver and not the file system overhead, even though that overhead is
> negligible.  Given such a configuration, comparing RAID-0 to RAID-5
> numbers shows the overhead brought on by the board and not the PCI bus
> nor OS.
> 
> I've been told by the manufacturer that my test is not valid, since
> it's not random (not reading and writing from random blocks) as normal
> use would be.
> 
> I'd like to know of what real performance numbers (using the "dd"'s,
> above) different RAID boards get.
> 
> Chris
> -- 
>   When I die, please cast my ashes upon Bill Gates
>    --for once, let him clean up after me!
>