On Sun, May 06, 2001 at 03:11:58PM +0200, Dominik Westner wrote:
> Actually there seems to be some difference between  the raid and an 
> internal disk.
> While the internal disk is locally much slower it performs better 
> through NFS compared to the raid.
> 
> Internal disk local performance:
> boo#  dd bs=1024k count=32 if=/dev/zero of=tmpfile
> 32+0 records in
> 32+0 records out
> 33554432 bytes transferred in 7 secs (4793490 bytes/sec)
> boo#
> boo# dd bs=1024k count=32 of=/dev/null if=tmpfile
> 32+0 records in
> 32+0 records out
> 33554432 bytes transferred in 1 secs (33554432 bytes/sec)
> 
> 
> Internal disk NFS performance:
> [icebox:/tmp/boo/tmp] root# dd bs=1024k count=32 if=/dev/zero of=tmpfile2
> 32+0 records in
> 32+0 records out
> 33554432 bytes transferred in 26 secs (1290555 bytes/sec)
> [icebox:/tmp/boo/tmp] root# dd bs=1024k count=32 of=/dev/null if=tmpfile2
> 32+0 records in
> 32+0 records out
> 33554432 bytes transferred in 1 secs (33554432 bytes/sec)
> 
> Is there a way to make the raid perform better through NFS? I mean for 
> NetBSD it's just some scsi device. I don't see how this could be tuned.

I suspect your RAID device doesn't like small transactions, it needs to
read/write full blocks several time, as NFS does fileops of size smaller than
a stripe.
Can you change the size of a stripe on the RAID device itself ?

--
Manuel Bouyer, LIP6, Universite Paris VI.           Manuel.Bouyer@lip6.fr
--