From: J. Andrew Rogers (andrew@ceruleansystems.com)
Date: Thu Sep 08 2005 - 16:14:55 MDT
On 9/8/05 1:37 PM, "Richard Loosemore" <rpwl@lightlink.com> wrote:>
>> Parallel == Serial == Parallel. It is not special, just complicated, and
>> fairly vanilla large systems design knowledge these days. It is complex,
>> but not *that* complex. I deal with this every day.
>
> Oh, spare me!! I'm sorry, but this is complete BS. If you deal with
> massively parallel programming every day, and you think it is no harder
> than anything else, then you have never seen it in your life.
>
> Not for a long time have I seen such a ridiculously naive dismissal of
> the problems of parallel programming.
What, specifically, is so difficult about parallel programming that is not
thoroughly discussed in transaction theory? Yes, it is more complicated,
but not prohibitively so. As I stated in the same email, most of the
complication is in trying to design languages/compilers that can usefully
hide it -- a very different issue. Actually writing massively parallel
system architectures to some set of constraints is not rocket science if it
is possible at all (and I have had to explain to suits more than once that
some things they wanted were not just difficult but theoretically
impossible). Maybe my lack of wide-eyed wonder is that while I do not do
much 'parallel programming' (whatever that means), I've been designing
massively parallel systems architectures with many different types of
constraints for years. There is no magic left in it for me.
Yes, I am dismissing the assertion that parallel programming is some kind of
arcane massively complex art because it is not. The tools may suck and many
people may not have much experience with it, but that is not the same as
being massively complex. The principles are fairly simple and the
engineering from those principles is straight forward.
And massive parallelism is rapidly becoming a standard required feature of
large-scale infrastructure systems. It would be hard to design these types
of systems and NOT be intimately familiar with the theoretical details of
massive parallelism.
J. Andrew Rogers
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:52 MDT