The comment "If you use RAID-5 you'll pay the price in performance" was made
in the context of a discussion of configuration options for the HP 12H disc
array.  You may have responded to a general discussion of RAID-5, but that
wasn't the discussion that was taking place.

The 12H array is, as the AutoRAID name implies, very much automatic.  You
can't really specify the RAID level.  All you can do is specify the usable
capacity of the array, and the controller divides the array's internal space
into RAID-1 and RAID-5 in whichever proportion is needed to provide the
requested capacity.  It automatically migrates data between the two areas,
keeping the more dynamic data in RAID-1 and static (and less frequently
accessed) data in RAID-5.  If you configure the capacity at a value greater
than that which would give you 100% RAID-1, you will experience a
degradation in performance.  Hewlett-Packard recommended an all-RAID1
configuration for our 12H arrays, and my experiments with higher capacities
confirmed the wisdom of their recommendation.

Regardless of any theoretical RAID performance characteristics and any
claims made by any manufacturer, I believe the best advice for someone
considering RAID is to get a demo unit of whichever product and see how it
performs on YOUR system with YOUR applications.

In re-reading the references in your earlier posting, those that addressed
performance advantages of RAID focused on increased transfer rates.  Seek
and latency times are, at best, the same as those for single discs, so the
overall theoretical improvement is small.  Disc arrays also promise to
improve performance through the use of caching.  The caching in arrays is
usually found to be of little or no value in improving overall system
performance in the case of MPE, because MPE already does an excellent job of
caching.  When you think about it, it makes sense that an operating system
can do a better job of caching than an external array, because it knows
about the data structures and how they are being accessed.  If a physical
I/O is required because the desired block of data is not in MPE's cache,
then it is unlikely to be in the array's cache either.  One hears of the
dramatic performance gains Unix systems experience when using arrays such as
EMC Symmetrix with a large cache, but those same gains are seldom, if ever,
realized with MPE.  Unix does not do the internal caching that MPE does.  If
you could flip a switch on an MPE machine that sys "I have an array with a
large cache, so don't bother doing caching", you could conceivably improve
performance by relieving the system of the burden and memory consumption of
caching, but I think that would only work in some corner cases, since MPE's
caching algorithm is so good.

-----Original Message-----
From: Steve Dirickson [mailto:[log in to unmask]]
Sent: Wednesday, December 08, 1999 8:49 PM
To: [log in to unmask]
Subject: Re: RAID5 Disc's


> First, I would point out that the concentration of discs on fewer channels
> is almost always one of the consequences of implementing RAID-5 in the
real
> world, so its effects are a valid issue.

I agree, it's definitely an issue. However, it does nothing to change the
fact that doing so and claiming that "RAID5 slowed my system down" indicates
confusion as to what really changed.

> Second, the web sites you pointed
> out compared RAID-5 favorably with other RAID levels for reads (mostly due
> to faster transfers, which is typically a small part of total access
time),
> but not with JBOD.

I don't think that's correct. I'm pretty sure that at least a couple of the
sites mentioned things like "read much better than a single disk, write
somewhat slower than a single disk."

> Since RAID-5 can be assumed to be somewhat slower than JBOD for reads,

Who assumes that? As shown on the sites you seem to have visited, RAID5 is
substantially *faster* than a single disk on reads.

> and lots slower for writes,

I don't think any of them said "lots slower."

> and since most systems have a
> mixture of both types of access, saying there is a
> performance penalty for
> RAID-5 is reasonable and accurate.

No, it isn't. Saying "RAID5 provides worse write performance than RAID0" is
correct, and I don't think anyone is claiming otherwise. Saying "RAID5
provides worse performance than a single disk", except in an extremely
write-intensive application, is neither reasonable nor accurate.

> Third, the question was made with respect to HP's AutoRAID product, which
based on my
> first-hand experience,
> suffers significant performance degradation when RAID-5 is used.  The
> degradation is greater for writes than for reads, but is
> evident for both.

Again, I can't really challenge other people's experience/benchmarking
without seeing it. But I can wonder, and ask, about the validity, i.e. was
it really a change to RAID5 from a single disk or was it a RAID0-to-RAID5
conversion? Of course, there may be issues with the AutoRAID product itself.
But my participation in the thread started in response to the statement "If
you use RAID 5 you'll pay a price in performance." Without qualifiers like
"If you use RAID 5 improperly..." or "If you use HP's AutoRAID product...",
that statement is simply not correct.


Steve Dirickson   WestWin Consulting
[log in to unmask]   (360) 598-6111