I have a question about disk and nfs I/O performance.  I'm using the pci ncr
driver with NetBSD 1.0.  I have a Quantum 1080s configured as sd0 that
goes like blazes (see below).  I've also got an ST12400N with a FreeBSD 1.1.5.1
ufs file system that has got real problems with writes.  With the
"bonnie" disk benchmark, using a 100 MB file, I get the following results:
(The hardware for all runs is identical: Intel Plato P5-90 mb, 64MB, ncr scsi)

              -------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
1.0**  sd0 100  3001 94.1  2998 21.4   632  7.4  2361 69.3  3152 23.2  58.0  4.2
1.1.59 sd0 100  2049 92.5  2117 50.0   583 11.0  1240 54.5  1272 16.7 121.2 10.8
1.0*   sd1 100   954 28.3   950  5.9   573  6.2  2153 61.7  2744 20.5  59.5  4.4
                 ^^^        ^^^
where the machines are

"1.0**  sd0" is NetBSD 1.0 ufs after "tunefs ..."              Quantum 1080s
"1.1.59 sd0" is Linux 1.1.59 ext2fs                            Quantum 1080s
"1.0*   sd1" is FreeBSD 1.1.5.1 ufs, running under NetBSD 1.0, Seagate ST12400N

The ST12400N had write performance similar to the very nice read performance
under FreeBSD 1.1.5.1.  Now I would just go and build a NetBSD ufs on sd1,
except for I noticed the following when running bonnie on an nfs mounted ufs:

              -------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
Net    sd0 100   134  9.5   137  1.8   109  3.5   697 40.2   759 14.1  32.7  4.7
                 ^^^        ^^^
Net -T sd0 100   158  9.7   157  3.0    39  1.8   521 26.9   656 12.4  30.0  5.7
                 ^^^        ^^^
1.1.59 sd0 100   238 26.9   260 20.1   127 14.3   224 15.9   250  8.4  21.4  9.6

Where the machines are:

"Net    sd0" is NetBSD 1.0, mount_nfs neighbor:/usr /u,      3c509, Quantum 1080s
"Net -T sd0" is NetBSD 1.0, mount_nfs -T neighbor:/usr /u,   3c509, Quantum 1080s
"1.1.59 sd0" is Linux 1.1.59, mount -t nfs neighbor:/usr /u, 3c509, Quantum 1080s

The server and client were configured identically for each test, i.e., NetBSD 1.0
on both, Linux on both.  I used an essentially quiescent coax network.
So, the questions are; are the low write performance/high read performance of sd1
and the nfs mounted drives related?  I don't see why they should be, but what the
heck.  Second, is there anything I can do to improve the nfs write performance?

Thanks,
Russell