Re: C++ inventor Bjarne Stroustrup answers the Multicore Proust
Questionnaire
On Sep 28, 6:31 am, "Chris M. Thomasson" <n...@spam.invalid> wrote:
"Chris M. Thomasson" <n...@spam.invalid> wrote in
messagenews:GxDDk.13217$VS3.2126@newsfe12.iad...
"Ian Collins" <ian-n...@hotmail.com> wrote in message
news:6k8eftF6e1smU1@mid.individual.net...
Chris M. Thomasson wrote:
"gremlin" <grem...@rosetattoo.com> wrote in message
news:v5KdncJt27DKykPVnZ2dnUVZ_oHinZ2d@comcast.com...
http://www.cilk.com/multicore-blog/bid/6703/C-Inventor-Bjarne-Stroust=
....
[...]
Q: The most important problem to solve for multicore software:
A: How to simplify the expression of potential parallelism.
Humm... What about scalability? That's a very important
problem to solve. Perhaps the most important. STM simplifies
expression of parallelism, but its not really scaleable at
all.
And what do you think simplifying the expression of potential
parallelism achieves, if not scalability?
[...]
Q: My worst fear about how multicore technology might evolve:
A: Threads on steroids.
Well, threads on steroids and proper distributed algorihtms
can address the scalability issue. Nothing wrong with
threading on steroids. Don't be afraid!!!!! I am a threading
freak, so I am oh so VERY BIASED!!! ;^|
I'm not too sure what Stroustrup was getting at here, but having
to write explicitly multithreaded code (with e.g. manual locking
and synchronization) is not a good way to achieve scalability.
Futures are probably significantly easier to use, and in modern
Fortran, if I'm not mistaken, there are special constructs to
tell the compiler that certain operations can be parallelized.
And back some years ago, there was a fair amount of research
concerning automatic parallelization by the compiler; I don't
know where it is now.
Of course, a lot depends on the application. In my server,
there's really nothing that could be parallelized in a given
transaction, but we can run many transactions in parallel. For
that particular model of parallelization, classical explicit
threading works fine.
--
James Kanze (GABI Software) email:james.kanze@gmail.com
Conseils en informatique orient=E9e objet/
Beratung in objektorientierter Datenverarbeitung
9 place S=E9mard, 78210 St.-Cyr-l'=C9cole, France, +33 (0)1 30 23 00 34