Subject: Re: ahc and raidframe questions
To: Curt Sampson <>
From: Greg Lehey <>
List: netbsd-users
Date: 06/27/1999 09:54:02
On Saturday, 26 June 1999 at 17:56:45 -0400, Curt Sampson wrote:
> On 23 Jun 1999, Chris Jones wrote:
>> Yeah, it's a good point.  But once the filesystem is there, it writes
>> files at about 1.7 MB/s.  This seems really slow to me.  Of course, I
>> might just have over-inflated expectations, but UW SCSI is supposed to
>> be fast.
> No, you don't have over-inflated expectations. I have an Adaptec
> 2940UW with a pair of new U2W Barracudas on it, striped with ccd.
> Bonnie writes at about 22 MB/sec and reads at 27 MB/sec.
> If you're interested in really good performance, RAID in software
> probably just won't cut it in the end. You can either mirror, or
> go with a hardware raid that uses tricks to get good performance.
> (Baydel arrays, for example, use RAID 3 rather than RAID 5, and a
> massive [64-128 MB] cache to turn the `lots of small reads and
> writes' I/O load into something that a RAID 3 can handle well.)

A lot of this depends on what you're trying to do.  RAID-3 is good for
things like video which require high transfer rates with sequential
access, but it won't help much for massively concurrent applications
such as web and ftp servers, since positioning takes too long.

I still don't have an opinion about the difference in performance
between "software" RAID and "hardware" RAID.  They're both software
RAID, of course: the real difference is just where the software gets
executed.  Conventional wisdom says that the parity calculations for
RAID bog down the CPU too much in "software" RAID, but in fact that's
seldom the case.  RAID-[2345] are all used when you're predominantly
reading, and when you consider the time it takes for, say, a RAID-5
write of 16 kB (about 20 ms), the 8000 xors you need to perform (in
the order of 100 Ás) don't make much difference.

Figures I've seen suggest that the DPT SmartRaid III and IV have
serious throughput problems on write.  If anybody has figures, or
wants to get them, I'd be interested in collecting them.  Again, I'd
plug my rawio program (, which
bypasses buffer cache and thus gives more accurate results for the
underlying storage equipment.

See complete headers for address, home page and phone numbers
finger for PGP public key