Proebsting's Law

[Jacqui Caren]

Dominic Mitchell wrote:
> This is all very well, but the majority of programs are i/o bound, not
> CPU bound.  So the optimisation is useful, but less important.  Unless
> you're doing mad simulations or calculating pi or searching for aliens
> or something.

My first job in IT was with a company that did ECAD logic gate level
board and circuit test simulations. I would try to debug core files fron
top end HP mainframes that had been running for up to three months on a
single board test model. I became very good at backtracking core dumps.
In this class of app I/O is never the problem - just memory and raw
computational power.

I remember when Cray first announced the M90 - the first machine able to
simulate a thermal+skin stress model of a 474. Up till then such models
could only run for aircraft parts such as tailfins or the wing.
Again I/O was not a problem. Memory and CPU speeds were the issue.

People do have a real commercial need for computational power unbound by 
I/O.

Jacqui