Are we talking apples to apples or apples to oranges here. I will not deny
the benchmarking done (sorry did not keep your e-mail) of how a RAID 5 is a
drag on performance, but I would like to see some specs. If I have 5 4GB
disc and 5 Raid5 arrays with 3 4GB disc, your telling me that the 5 4GB
configuration is going to beat it in a read test?
When I look at the specs for RAIDs I see people running out of I/O slots and
replacing a 4GB or 9GB with a single RAID of 3 to 5 9 or 18 GB drives. Now
let us look at this logically. System with 9 4GB drives is swapped out for a
single RAID5 with 3 18GB drives. IT says great we took 9 I/O slots and used
only one and we still have 36GB usable space, great. But you took 9
independant, meaning each being able to respond to a request, disc
mechanisms and put one in. The throughput is great on the RAID, but that is
for only single large data reads. The transient (right word?) time for each
I/O is not that much faster on a RAID that a single disc. (Time from when
request is made by the CPU til the disc responds) This is not considering
cache.
Hopefully you can see my point. Because of the cost of the RAID solution,
the disc which populate the controller are usually maxed out to help justify
the cost. But no way can I see where is a read situation that 5 disc and
beat 5 RAIDs. Of course you will say that that is not fair because I am
looking at 5 disc to minimum of 15 disc. Great and neither are you being
fair to take 15 disc and compare it to a single 15 disc RAID.

James

-----Original Message-----
From: HP-3000 Systems Discussion [mailto:[log in to unmask]]On
Behalf Of Lee Gunter
Sent: Wednesday, December 08, 1999 10:29 AM
To: [log in to unmask]
Subject: Re: RAID5 Disc's


Agreed ... that happened to us.  What RAID5 *did* do for us was to increase
the
system's stability -- we never experienced an outage due to a hard failure
of a
RAID5 HDA.  Also, at the time we decided to go with it, we needed to reduce
the
overall equipment footprint on the floor, and this was the most compact
technology (GB/sq. ft) available for MPE/iX in 1993, IIRC.
----
Lee Gunter                    The Regence Group
Supervisor, TRG HP/MPE Systems     503.375.4498
----
Opinions expressed are solely mine.




From: Bill Lancaster <[log in to unmask]> on 12/07/99 06:55 PM

Please respond to Bill Lancaster <[log in to unmask]>


To:   [log in to unmask]
cc:    (bcc: Lee Gunter/BCBSO/TBG)
Subject:  Re: [HP3000-L] RAID5 Disc's




I have to disagree.  I've never seen RAID 5 increase anyone's HP 3000
system performance.  In *many* cases I've seen RAID 5 seriously hurt
performance.  In many of those cases, going back to RAID 1 or JBOD in turn
increased performance.

Bill

At 04:29 PM 12/7/99 -0800, Steve Dirickson wrote:
> > over RAID level in these arrays.  If you use RAID 5 you'll
> > pay a price in performance.
>
>Probably not. RAID5 has slightly worse write performance than a single
>spindle (due to the need to update the parity information), but
>substantially better read performance. Since reads outnumber writes by an
>order of magnitude or two (or three...) in "normal" installations, RAID5
>should *increase* system performance.
>
>Simple explanations/discussions:
>  http://www.digidata.com/raiddesc.html
>  http://www.acc-sd.com/site/raidlevels.htm
>  http://www.adaptec.com/technology/whitepapers/raid.html (note the comment
>that "Database servers are an example" of where RAID5 works well)
>
>More extensive discussions:
>  http://www.eurologic.com/tn/tnwp2.htm
>  http://www.acnc.com/raid.html
>
>
>Steve Dirickson   WestWin Consulting
>[log in to unmask]   (360) 598-6111