Re: USII MP (Re: USIII/280R)

[Chris Ross <cross+netbsd%distal.com@localhost>]

On Jan 4, 2011, at 07:58 , Julian Coleman wrote:
>> My Enterprise is a 6-way box and I remember some people claiming that they 
>> had no trouble with dual-cpu systems at the time, but killing the 6-cpu 
>> machine was dead easy. Just run build.sh and wait a couple of hours. Or 
>> run build.sh -j7 and wait less than 9 minutes. But maybe those problems 
>> are somehow specific to old SBus Enterprise boxen... :-/
> 
> I installed a Dec 17 current on my 4 processor E420R (450Mhz USII) and tried
> a build.sh -j 8 (distribution + X + kernels, src and obj on NFS).
> 
> With 5.* and earlier, it would always hang at some point.  With current, it
> finished the build with no problems (9 hours for distribution + X, 20-25
> minutes per kernel).  So, I would say that USII MP is much improved in
> current.  I haven't tried with more than 4 CPU's though, but I think that
> it's worth trying again.

  That's interesting, as I've been running an E420r (4x450 Mhz US-II) as a 
production machine, running the netbsd-5 branch, from shortly before the 5.0 
release.  There were a small number of situations a year or more ago where it 
would panic and lock-up failing to dump/reboot/etc, but I think those (the 
panic part I mean) all were assumed to be a "threading doesn't work [well] in 
bind" problem.  I have turned off threading for bind in my source tree, but 
other than that, I think I'm running a standard build.  And, it "just works" 
for me.  I don't do full OS builds on it, as it's a lot slower than my x86 
build machine, but I build the pkgsrc pieces I need without any significant 
problem, and it acts as my sendmail/uw-imapd/apache22 machine, albeit lightly 
loaded, with no problems.

  This makes me want to check out a copy of the sources and try a full build, 
just to see if it really is as stable as I think it is, or if I'm just lucky.  
;-)

                        - Chris