On Mon, 18 Oct 2004, Manuel Bouyer wrote:

> Date: Mon, 18 Oct 2004 23:17:15 +0200
> From: Manuel Bouyer <bouyer@antioche.eu.org>
> To: Karl Janmar <karl@utopiafoundation.org>
> Cc: tech-kern@NetBSD.org
> Subject: Re: ATI IXP IDE chipset support
>
> On Mon, Oct 18, 2004 at 11:07:18PM +0200, Karl Janmar wrote:
> > > these are really poor performances, something is wrong.
> > > >From a seagate 8G drive (ST38410A) connected to a i810 controller:
> > > rochebonne# /tmp/tst /dev/rwd0d 3000
> > > 4577374 us, 40.962351 MB/s
> > >
> > > >
> > > > these results are a little bit strange(for me as I am not an expert),
> > > > because  I get this when using dd, /mnt is on wd0e (the ATA disk):
> > > > # dd if=/dev/zero of=/mnt/apslask bs=4096 count=100000
> > > > 100000+0 records in
> > > > 100000+0 records out
> > > > 409600000 bytes transferred in 17.693 secs (23150398 bytes/sec)
> > > >
> > > > This doesn't ofcourse exacly test the same thing but....
> > >
> > > This is a 400MB file, if I got it right. Could it fit entirely in RAM ?
> >
> > I got 512MB of RAM in that machine, so yes?
> > But if I issue a sync(8) it doesn't delay more then 1/2 second or so.
> > Doesn't that mean that the data found it's way all the way to the disk
> > cache?
>
> Yes, it should. Hum, could you try reading a file (after making sure it's
> not any more in cache) ? Myabe it's slow in only one direction ..
>

I did the read test with dd now, first making sure that the file wasn't in
the cache.

# dd of=/dev/null if=apslaskr bs=4096 count=100000
sync100000+0 records in
100000+0 records out
409600000 bytes transferred in 14.996 secs (27313950 bytes/sec)

I could also se on sysstat vmstat that I hade about 25M of disktransfer
from wd0, as expected. So it seems that there is no problem to read/write
to the blockdevice on my disk... but the character device....

- Karl