Stoned koala bears drooled eucalyptus spit in awe as Havard Eidnes 
exclaimed:


> FWIW, I also have a CS20 which I've been running -current (currently
> 1.6K) on for a while, and I've been running it in MP mode with no
> observable major problems (OK, I'll admit that I've managed to wedge
> it once to the point of it requiring a power-cycle to recover; I think
> the problem involved ktrace, but I did not pursue this further).

I've got an AS4100 with quad 300 mhz processors and I'm running into 
problems similar to Dave's.  The machine can run under a heavy CPU load for 
forever with no problems, but heavy I/O brings it down.  Particularly if 
it's done over NFS (As a server, haven't tested as a client).

> Are you trying to use any particular functionality in the NetBSD
> kernel?  Raidframe, perhaps?  I'm not using raidframe, and if I recall
> correctly, there have been locking issues in the raidframe code which
> have not yet been fully resolved(?), and it's not inconceivable that
> these problems could cause more severe problems on MP systems.

I've got RAIDFrame compiled into my kernel but not using it for anything 
right now.

> I'm assuming as a matter of course that you're running -current on the
> systems?  There has gone in machine-dependent fixes to the alpha port
> related to MP functionality after 1.6 was released.

I'm running 1.6K.  1.6-stable was far worse for NFS on an MP machine. It'd 
barf at the first sign of NFS activity, with lots of debug messages that my 
feeble brain cannot understand. However, with 1.6K, it just locks up hard, 
requiring a reboot, without any kind of logging or error message.

-Jeff



--
Jeff Workman | jworkman@pimpworks.org | http://www.pimpworks.org