On Sun, Oct 05, 2003 at 08:01:04PM +0200, Rodrigo Fernandez-Vizarra wrote:
> Hi,
> 
Hiya,

> Ian, we are not the only ones with this problem...
> 
Phew - that's good to know.  I was starting to worry about the hardware
and/or my load of NetBSD.

> I'm trying to develop some test to see if its only a network problem or 
> is a network + hd i/o problem. Any kernel developer in the list?
> 
Interestingly, I've managed to transfer 3.5GB of data in a single ftp
session without any problems.  In fact, that's the main problem I have in
tracing the issue - it's just *too* stable for me.  I have a moderate amount
of RAM but still tend to hit swap in under 36 hours but my system doesn't
fall over until 23 days or so, which means that my physical RAM ought to be
good.

The network I/O is poor, and I've spent ages messing with the different PIO
and DMA modes in custom kernels and have only managed to slow the system
down compared to the conf/DEFAULT settings.  I haven't messed with the
nmbclusters option (mentioned in the archive) but I am wondering if my
setting the kern.maxvnodes to 40000 has anything to do with the crashes.
The vnodes are exhausted within a few hours, but in doing so have I taken
the value too high ?  No idea.

I have now reorganised my system and have found enough parts to get a
PIII 450 up to take the place of the Qube.  That'll take me a couple of days
to sort out and test a NetBSD/x86 install but if I can then simply mount the
cobalt drive then I'll have the Qube free (for the first time in over two
years :) so I can at least reboot at will and keep a serial cable on for
kernel debug.

I might dig out the Linux restore CD and see what the disc and network
throughput is on an original setup, just to give a baseline for the hardware
to live up to.

TTFN,
-- 
ian.
PGP Fingerprint: D170 35A3 C858 6E85 9B5B  1557 4CD5 6F6F E176 2D0A