> 
> Guess what, the thing ran 5 hours faster under SparcLinux compared to
> Solaris2.5.1, about 5 1/2 hours faster than SunOS4.1.4.  Let's say he
> needs to get as many jobs like this as he can done every week, I think
> it matters to him the things I have presented.   He's told all of the
> ARPA speech processing people upstairs, and they are all asking me
> when they can get cycles on my SparcLinux machine so they can be more
> sure to meet their deadlines.  Thank you very much.
> 
5 hours out of 96.  And then somebody looks at his code once he's
published a paper based on a year's worth of computations and finds
he's accidentally reversed a sign in a formula and he has to start
again (I've actually seen this happen a few years back).  Nevermind
the type of optimisation that using a more appropriate algorithm or
programming language might be able to offer.

I believe in squeezing every bit of performance one can out of a
computer, but I'd rather use Plan 9 where cross compilation is a given
and (partial, possibly incomplete) optimisation is in the loader than
have to adjust for each different architecture I manage.  Userland is
all very well, but I spend most of my time _managing_ networks and get
little choice as far as architecture goes.

>    Most modern machines spend most of their time IDLE.
> 
> Bwahaha, many people would beg to differ with you highly.
> 
Your workstation is probably idle right now, while you read this.  Any
optimation in Linux that can make your CPU cycles available to those
in your organisation that might be needing them right now?

Have a look at how Bell Labs (now Lucent) addressed that type of
problem (http://plan9.bell-labs.com/plan9/), then tell me how well
your efforts compare with theirs, specially in the realm of
portability.

-- 
Lucio de Re (lucio@proxima.alt.za) -------------------------------------
-------------- 'Cancel my subscription to the Resurrection.' (The Doors)