> > > It's possible that NetBSD doesn't use DMA for you disk on the tower.
> > > Can you post your dmesg to see what chipset this is exactly ?
> >
> > Sure.  Here are ~20 relavent lines from near the top.  (The rest can be
> > provided if really desired.)  Note that it claims to be using DMA.
> >
 [...]
>
> Hum, OK. 1.5.x has problems with this chipset which may explain your
> slowdowns. This should be fixed in -current, but back-porting it to 1.5
> involves more than just pciide. I'll try to get something done, though.

Note that GNU/LINUX has the same performance problems.  (Or rather, Debian
GNU/LINUX from about a year ago also had abysmal dd read performance.  I
didn't run bonnie++ from there, though.)

I don't know if that affects your estimation of whether -current will fix
my problem.  However, I can try booting a somewhat more recent -current
(sources from no earlier than Dec. 1 of 2001, though I think that I did
one more update of the sources on that machine---or I can grab a snapshot
and drop its kernel in).  If -current is stable enough at the moment, it
might be worth doing anyway.  (Lennart Augustusson(sp) has made some
changes to USB uhid handling that I'd like to have on my main box; I've
been reluctant to commit the machine to -current, but this might be enough
added argument to doing so.  (^&)

I'll try to remember to try that later tonight.  Let me know if there's a
particular cut-off date I should have in mind w.r.t. which version of
-current to use.


  ``I probably don't know what I'm talking about.'' --rauch@math.rice.edu