Is concurrency hard?

[Ulf Wiger]

Den 2005-11-02 18:20:29 skrev David Hopwood  
<david.nospam.hopwood@REDACTED>:


> But enough of the armchair physics. I'm sure we're giving any real
> physicists in the audience fits.

I'm sure of it. (:

Going back on topic, I spent some time following the attempts of
companies like Encore and Expertelligence in the early 90's to
deploy software on highly parallel architectures (in the case of
Encore, Unix machines interconnected with very fast pipes,  and
in the case of Expertelligence, Lisp compiled down to transputer
arrays.) Back in those days, it seemed too difficult to make the
mental shift from sequential programming to multipro. Encore had
to rewrite software quite extensively in order to port it to
its architecture - AFAIR, they thought many of the rewrites could
have been avoided if the original programmers had followed some
common sense rules. But since deployment on distributed architectures
wasn't really conceived by most at the time, this didn't happen.

Most attempts to exploit concurrency in hardware seem to have been
severely limited by what software designers were ready/willing to
cope with. The current shift to multicore is driven by the apparent
fact that we've exploited all other known alternative routes to high
performance. Time then to force programmers to re-think. Luckily
for Erlangers, the mental shift shouldn't be that traumatic.

/Uffe
-- 
Ulf Wiger