HP3000-L Archives

January 2000, Week 3

HP3000-L@RAVEN.UTC.EDU

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Chris Bartram <[log in to unmask]>
Reply To:
Date:
Thu, 20 Jan 2000 13:09:43 -0500
Content-Type:
Text/Plain
Parts/Attachments:
Text/Plain (39 lines)
 In <[log in to unmask]> [log in to unmask] writes:

> Andreas Schmidt wrote:
> > So, from our perspective, we can say
> > * Mirror/iX is a cheap solution compared with Disc Arrays to ensure a high
> > availability on business critical servers,
> > * the SCSI drives & interfaces are very reliable,
> > * the repair process (if needed) is proven and straight forward.
>
> This exactly matches my former customer experience with Mirror/iX at cccd.edu.

(speaking for a site I manage)
We're somewhere in between. We've lost about 8-10 mirrored discs in the past
year (out of a set of 18 pairs of 4.3Gb drives), but have yet to require a
reload (i.e. no lost data on the mirrored discs). Mirror/iX does seem to
"agressively" flag discs as bad, and we've had some problems with discs
dropping offline (due to what HP suspects is excessive i/o rates - probably
due in part to some mirrors not being on separate channels... a problem we
recently discovered on a system that HP (mis)installed).

We've also lost 3 of the same model discs on non-mirrored volumes in that same
period; 3 reloads of those volume sets.

It does seem to reduce i/o loads; our 997-400 runs VERY high i/o rates, and
I suspect we'd be seeing serious performance problems without it.

Overall, it seems to help our performance a bunch; we haven't lost ANY data
(despite all the dead discs), but had a LOT of false alarms. All told I'd
have to recommend it (much better than no mirroring, and splitting reads
across mirrored pairs helps i/o throughput alot).

That said we also bought an XP256 disc array (though pricetags on those
beasts are not for the faint of heart - 1Tb was about $1M) and are experiment-
ing with that as well. We were *extremely* disappointed to find out that the
100% redundancy on those units doesn't apply to connections to HP3000s, which
don't allow redundant/failover paths to the array. So much for the 5 9's.

   -Chris Bartram

ATOM RSS1 RSS2