Hi,

	I was wondering if there are hints anyone can provide on how to
increase the write performance of RAID-5 devices configured with RAIDframe
on NetBSD-1.5.3_ALPHA.  I've fiddled with different block sizes and
cylinder counts when newfs'ing, but write performance always roughly an
order of magnitude slower than read performance.  My setup is 3 SCSI Ultra
160 drives on one AIC-7899 controller, and there are 2GB and 15BG partitions
on each drive devoted to two separate RAID-5 partitions.  The tests I'm
running are currently just on the RAID-5 composed from the 2GB partitions.
Typical results from bonnie++ are:

Version  1.01       ------Sequential Output------ --Sequential Input- --Random-
                    -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
flatland       300M  4855  10  4833   2  3398   5 30301  94 37835  18 163.3   2
                    ------Sequential Create------ --------Random Create--------
                    -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
                 16    39   6 +++++ +++   162   3    39   6   952  99    75   4

(using 32 sectors per stripe unit, 8K block sizes, and 16 cylinders per group).

I know it's not a fair comparison, but for the sake of having something
concrete to compare to, the results for a hardware RAID-5 device on a slow
alpha are:

Version 1.02a       ------Sequential Output------ --Sequential Input- --Random-
                    -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
hardware-raid5 300M 16281  51 15839   9  7809   7 16598  64 16804  10 188.7   1
                    ------Sequential Create------ --------Random Create--------
                    -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
                 16    80   8 27481 100   159   4    82   8  1248  99   175  11

Any hints on improving performance will be much appreciated.

	Thanks,

	-- Johnny Lam <jlam@jgrind.org>