> I've been following this discussion of parallel programming for some
> time. You don't seem to be asking yourselves: is this primitive
> useful for writing any real algorithms?
Okay.
> Try this: write a parallel matrix multiplier using DSM, then again
> using HPF. The HPF version will turn out both shorter and faster.
Never done it, but ill take your word for it..
> Then, write a parallel AI search application using DSM, then again
> using parallel prolog. The prolog version will turn out both shorter
> and faster.
Again
> Then, write a parallel N-body simulation using DSM. Then write it
> again using charm++. The charm++ version will be both shorter and
> faster.
Again.
> I would challenge you to find ANY problem that can be implemented in
> DSM, for which I can't find a better construct.
I challenge you to find any solution for which I couldn't find a problem
that was better solved differntly.
> You might argue that DSM is useful for implementing HPF, prolog, and
> charm++. But it's not true. We have implemented object-oriented
> languages using DSM, and we have implemented them using RPC, and the
> RPC is faster. We have implemented HPF using DSM, and we have
> implemented HPF using explicit generation message-passing code, and
> the latter is faster.
Faster on what? Certantly not faster on a smp computer.
Give facts, tell me *what* it was faster on.
(hey, and how about some URLs for the stuff you mentioned above, I havn't
heard of all of it).
> - Josh
Gregory Maxwell
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu