Thinking about this whole more-workstations-than-is-remotely-sensible
thing, a couple of points occured to me. Firstly, with it getting to the
point where whole installations of Sun3 gear are being junked it has
become feasible for an individual to amass large numbers of the things.
Secondly, with the reusability of storage media and the networked design
of the range, odds are that these will be mostly diskless. Thirdly,
given that monitors tend to die a lot quicker and are, on the whole,
more expensive to replace, then it's also likely that many of these
easily-acquired systems will be in the form of lone CPUs. So, what can
we do with these? It seems a shame to waste such a lovely resource, but
just slapping a serial terminal on and treating them as text-only
workstations doesn't quite seem to cut it.

Then I thought, what are the things we like about our Sun3s and what are
the things we *don't* like, and the answer seemed obvious. We like their
big monitors, we like X, we like BSD, we like the way they love to network
and we like the way they smell. 
What we don't like is the fact that if you try to do much anything other
than run X it can take a long time. If you try to compile your kernel it
takes a very long time. If you try to do the whole source tree then you
better have another machine on that afore-mentioned network, because the
one you're using will *not* be available for some time.

Anyway, the solution...

Given that it's not so difficult to get yourself a bunch of Sun3 CPUs
all networked together and sitting in an aesthetically pleasing pile,
might it not be possible to implement some sort of load-balancing
scheme? After all, it's an almost perfectly homogenous processing
environment so it should be reasonably simple to automate the
distribution of the workload even if only in a pretty huge-grained sort
of way. It should at least be possible to spread the compilation task
around - how about it Rick, volunteer to be the official port-sun3
kernel compiler?

Now that I think about it, doesn't GCC already support some sort of
compilation in multi-processor environments? How much modification would
it take to make use of that?

I realise, of course, that the sensible answer is and has always been
'stop mucking about and just buy a new machine instead of spending a
fortune in man-hours on that ageing piece of junk' and that any scheme
like that described above is in essence pointless. On the other hand,
people who say things like that have no soul. I *want* a huge pile of
workstations, I don't care if their total processing power is still less
than that of your wristwatch, it's my house and I'll do what I like.

Neal.