Hello all,
I've made some series of bonnie runs for RAID-0. You can look at the results at
http://plasma-gate.weizmann.ac.il/~fnevgeny/tmp/raid0.gif.
* The best block read performance (that's what I'm mainly interested in) is for
32k or 64k chunk sizes.
* 32k seems also like a "magic" number for which all (except char-based) I/O
parameters are enchanced equally.
* 256k is best for applications havily using *seek()
* write rates: for my specific case, there is a saturation due to (I believe)
the SCSI controller (40MB/s), but for faster buses or multi-channel or
multi-adapter configuration one would probably get the theoretical 3-fold
performance increase.
* I still don't understand why the read performance is ~ 30% worse than it
should be
Note: I've tried all kinds of ext2fs options (-R stride=1,2,4,8,16,32,64,
-g 7936) with no noticable effect.
Regards,
Evgeny
--
____________________________________________________________
/ 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 >=====================================+ |
|______________________________________________________________|
