Hello all,
I need a very fast device for reading/analyzing huge chunks of data, typically
hundreds of MB (which are created by a slow process, so write performance is not
really important for me). From what I've read in the HOWTO, RAID-1 with N disks
should give an N-fold increase in the read performance; plus, the redundancy, to
avoid doing a lot of backups - this seemed exactly what I needed. Well, RAID-0
might be an option, too. However, the benchmarks disappointed.
OK, first, the hardware configuration:
Dual PII 450, 512MB RAM, aic7895 on-board (dual channel, 40MB/s) controller.
Disks: a 1GB disk on channel A (for OS itself), 3 wide 7200 RPM 4.3GB
(Barracuda) on channel B (to form the RAID array(s))
Each of the three disks (sdb, sdc, sdd) are partitioned as following (not final,
just for testing):
1 128M Linux swap
2 128M Linux swap
3 1GB Linux native
4 1GB Linux native
Software: kernel 2.0.36, raidtools-0.42 (I've also tried raidtools-19981214-0.90
with raid0145-19990108 patch, with practically no performance improvement).
Output of /proc/mdstat:
Personalities : [1 linear] [2 raid0] [3 raid1] [4 raid5]
read_ahead 120 sectors
md0 : active raid0 sdb3 sdc3 sdd3 3084288 blocks 256k chunks
md1 : active raid1 sdb4 sdc4 sdd4 1028096 blocks [3/3] [UUU]
md2 : inactive
md3 : inactive
The filesystem on md0 is created by "mke2fs -b 4096 -R stride=64"
hdparm results:
# hdparm -tT /dev/sdb #(raw disk; same for sdc and sdd)
/dev/sdb:
Timing buffer-cache reads: 128 MB in 1.29 seconds =99.22 MB/sec
Timing buffered disk reads: 64 MB in 4.46 seconds =14.35 MB/sec
# hdparm -t /dev/md0 #(raid0)
/dev/md0:
Timing buffered disk reads: 64 MB in 2.90 seconds =22.07 MB/sec
# hdparm -t /dev/md1 #(raid1)
/dev/md1:
Timing buffered disk reads: 64 MB in 11.81 seconds = 5.42 MB/sec
"iozone 800" on /raid0:
35469801 bytes/second for writing the file
19315238 bytes/second for reading the file
"iozone 800" on /raid1:
7901109 bytes/second for writing the file
9720287 bytes/second for reading the file
"bonnie" gives similar results.
-----------------------------------------------------------------------------
The bottom line:
RAID-0: write - OK, read - half of expected
RAID-1: complete disaster. write - about half of expected, read - 3-4 times
worse than expected.
-----------------------------------------------------------------------------
Any hints?
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 >=====================================+ |
|______________________________________________________________|
