Martin Weinberg wrote:
>
> Is this performance difference expected or did I make an mistake
> somewhere?
Looks sane to me. The bad news is, RAID 5 hurts. :) The good news is,
RAID 5 doesn't hurt as bad as a failed disk in a RAID 0 array.
I have an Ultra160 SCSI RAID system with 3 drives and I'm getting these
results under reiserfs:
> Kernel 2.2.14-5.0, Hardware RAID 5, reiserfs, 32 KB stripes:
> > -------Sequential Output-------- ---Sequential Input-- --Random--
> -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
> Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
> permethr 1000 4491 28.7 4307 11.8 3295 7.7 14573 52.6 28986 21.0 266.3 4.7
> Kernel 2.2.14-5.0, Software RAID 5, reiserfs, 128 KB stripes:
> -------Sequential Output-------- ---Sequential Input-- --Random--
> -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
> Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
> permethr 1000 8934 56.0 10373 26.7 5584 12.7 13565 48.8 26665 18.7 305.5 5.1
Note that in most respects software RAID 5 is much faster than hardware
RAID 5 on this machine! (256 MB SDRAM, Celeron-500) I'm using the
Adaptec 2100S board, FWIW. I'm still using the hardware RAID, though,
because a) I'm paranoid about software RAID 5, b) our application is
mostly linear reads, where the hardware RAID has a slight edge, and c)
we can't afford to spend CPU power on RAID 5 calculations in this
system.
If I had 8 drives like you do, I could outperform your system, but I'd
expect that from SCSI over IDE...
Here's some more interesting data, hardware RAID 0 vs. software RAID 0
under ext2fs on the same machine:
> Kernel 2.2.14-5.0, Software RAID 0, ext2, 128 KB stripes, stride=32:
> -------Sequential Output-------- ---Sequential Input-- --Random--
> -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
> Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
> permethr 1000 17154 79.8 17827 16.9 10073 21.1 21526 74.8 44186 29.3 316.7 4.1
> Kernel 2.2.14-5.0, Hardware RAID 0, ext2, 128 KB stripes:
> -------Sequential Output-------- ---Sequential Input-- --Random--
> -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
> Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
> permethr 1000 12366 57.6 15163 14.0 6717 13.3 14889 50.3 16132 8.4 67.3 0.6
Again, software RAID 0 is faster on this system than Adaptec's 2100S
hardware RAID, but with even better margins this time. Too bad we can't
get away with RAID 0 for this project....
--
= Warren -- ICBM Address: 36.8274040 N, 108.0204086 W, alt. 1714m
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to [EMAIL PROTECTED]
