On 4 Mar 2002 00:15:01 -0600, Andy Ball wrote:
> 
> Hello Kent,
> 
>   K> I have a 2940UW and as far as I can determine, under the best
>    > conditions with NetBSD 1.5.x, it can't achieve even 20Mb/s in
>    > benchmarks with (three different) scsi2wide drives that should be
>    > able to approach 40Mb/s.
> 
> I think Fast, Wide SCSI-2 would be limited to 20Mb per second 10MHz x
> 2 bytes per transfer absolute tops.  This is probably academic anyway
> since a single drive isn't likely to sustain enough throughput to
> saturate the SCSI bus (it has to sit and wait for the data to fly
> under the head).

Here's what bonnie reports with 2940UW and three IBM DDRS-34560's :

1.4G athlon sd0e ibm 4.5G scsi-2:
     -------Sequential Output-------- ---Sequential Input-- --Random--
     -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
  MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
 100  8698 10.5  8532  2.4  2883  1.2 10834 19.3 11974  3.4 110.0  0.9

1.4G athlon ccd0a (2 striped) ibm 4.5G scsi-2:
     -------Sequential Output-------- ---Sequential Input-- --Random--
     -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
  MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
 100 12981 17.0 12421  3.5  4049  1.8 10570 19.1 11252  3.3 102.9  0.9

>   K> Since the same machine runs a 7200 ata100 at 40Mb/s (even writes)
>    > with the same benchmarks, can I assume a problem with NetBSD and
>    > the Adaptek firmware (even updated to the latest version) or
>    > hardware?
> 
> I don't know how much of a role the Adaptec firmware plays once BSD
> has booted.  In any case, you can't expect Ultra throughput from
> SCSI-2 devices.

I have no idea what one should be able to expect. The Adaptek
literature and the IBM drive literature claim theoretical 40Mb/s.
I would expect that under optimal conditions, running a benchmark
such as bonnie, one should be able to do better than 12981K/sec
with a pair of striped drives though, especially when it ran faster
with a 400MHz K6-2:

400M K6-2 sd0e ibm 4.5G scsi-2:
     -------Sequential Output-------- ---Sequential Input-- --Random--
     -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
  MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
 100 12101 64.1 11702 19.9  2968  6.1 12707 59.9 12781 13.1 112.1  1.8

400M K6-2 ccd0a (2 striped) ibm 4.5G scsi-2:
     -------Sequential Output-------- ---Sequential Input-- --Random--
     -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
  MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
 100 15688 85.3 15252 27.4  4266  8.2 10796 51.3 11186 11.1 110.4  1.9

even though %cpu is way up. I still suspect there is an interrupt
timing problem with the faster cpu in the NetBSD driver for this.