"m. allan noah" <[EMAIL PROTECTED]> wrote:
> actually did not save any hard numbers, but on a 3 disk setup i had here,
> raid1 was faster under bonnie with the test size 3x my ram.
What mke2fs parameters did you use? And which chunk size (for RAID-5)?
> less than
> that, and i found the results pretty variable.
>
> but, our boy was talking about a single hotel firewall, and yet this
> discussion now revolves around beowulf clusters :)
Yes, sure, the performance is probably not an issue there. But since the
perfomance topic was touched, I wanted to know whether somebody has actually
seen the theoretical N-fold read enhancement of an N-disk RAID-1 array. I
couldn't, though I did some more or less extensive tests with different
parameters.
If anybody interested, I put a comparison of software RAID 0, 1, and 5 at
http://plasma-gate.weizmann.ac.il/~fnevgeny/tmp/raid.gif. The chunk size was
32k (for 0 and 5), fs block size 4k and for RAID-5, in addition, partitions
were formatted with "-R stride=8 -g 7936" (0 and 1 seemed insensitive to
anything but block size).
Except for the bad performance of RAID-1, a very surprising fact is that
RAID-5 actually make significant improvements (as opposite to what said in
the HOWTO). I'd find an explaination for an N/(N-1) read performance
improvement, but how it happened for _writes_?!
Regards,
Evgeny
PS. The box benchmarked is a dual PII-450 with 512MB RAM, the disks are 4.3G
7200RPM wide barracudas attached to a channel of the on-board adaptec 7895,
partition size - 1GB
--
____________________________________________________________
/ Evgeny Stambulchik <[EMAIL PROTECTED]> \
/ Plasma Laboratory, Weizmann Institute of Science, Israel \ \
| Phone : (972)8-934-3610 == | == FAX : (972)8-934-3491 | |
| URL : http://plasma-gate.weizmann.ac.il/~fnevgeny/ | |
| Finger for PGP key >=====================================+ |
|______________________________________________________________|