[ On Friday, March 11, 2005 at 18:43:59 (+0300), Odhiambo G. Washington wrote: ]
> Subject: [exim] Storage systems - again!
>
> Never mind the cost, because it's actually what has driven me to this combo.
> It's a fraction of what NetApp, IBM, etc offer, as much as I can see.
Indeed Xserve RAID arrays offer one of the most cost-effective mid-range
performance storage solutions.
> XServe RAID uses Ultra ATA disks. These are not SCSI disks, nor SATA
> disks.
Totally irrelevant (except for the fact this seems to be where the cost
savings comes from). As long as it goes, and goes as fast as it should,
and meets your reliability requirements, that's all that matters.
> It's a RAID system anyway and I think my choice of RAID
> architecture will be most important.
Yes, but....
I would stay very far away from the xSAN part. It's a lot of very
complicated and easy to break additional software for what's likely a
very tiny benefit.
Just get yourself a nice 2Gig FibreChannel switch, plug everything into
it, and configure the Xserve RAID box(es) to provide uinique LUNs for
each partition, sized as required.
If you really need to concatenate, or mirror, partions then just do it
at the OS level -- all modern OSs have at least the basics needed for
doing this (e.g. a CCD driver, or RAIDframe, etc.). Indeed this is how
Apple themselves propose getting speed and reliability by mirroring both
sides of the Xserve RAID box in the OS.
It doesn't sound like you really need any of the other xSAN stuff or to
share partitions between multiple hosts simultaneously and I would
really strongly suggest avoiding digging yourself into this hole if
possible.
It seems the Xserve RAID with a fully configured cache RAM option and a
full compliment of disk spindles, can perform very well with full RAID-5
partitions on each side (they actually consist of two RAID controllers
with the disks split between them, sharing only passive components).
I've got a similar configuration about to go into production at a client
site. It's just one Xserve RAID shelf, fully configured, with one host
connected by a pair of Qlogic FC host adapters. I'm splitting the two
RAID-5 groups into four LUNs (mostly to avoid the 1TB "spindle" limit
imposed by my host OS (NetBSD/alpha) disklabel). If/when we need to
expand the capacity we'll just add another shelf and an FC switch as I
suggest above.
(I'd really like to get another identical shelf and cross-mirror the
partitions between them, but that's too much expense for too little gain
for the application!)
--
Greg A. Woods
H:+1 416 218-0098 W:+1 416 489-5852 x122 VE3TCP RoboHack <woods@???>
Planix, Inc. <woods@???> Secrets of the Weird <woods@???>
