The status of parallel programming languages

[r***@gmail.com]

Hello,

I was looking around for languages for parallel and distributed
programming, and I have mostly discovered ruins. Eg There used to be
something called CxC and the company changed direction, there were
some systems like Concert, Code http://www.cs.utexas.edu/users/code/
and many others I didnt bother jotting down than have seen no active
development in years or decades. Correct me if I am wrong, it looks
like all the effort is in libraries like MPI and Jini, which is not
the same thing. I think the only project which shows some vitality is
Cilk. Other dead bodies recovered include some distributed Java JVMs
and distributed Python. The most interesting alternative and antidote
to this trend seems to me to be http://boinc.berkeley.edu/

Have I missed something big? Is today's scalability and "speedability"
option simply the conversion and distribution of tasks into batch jobs
(as in Sun's grid) or web service calls (as in Amazon''s cloud)?. I am
guessing distributed programming proved to be too hard in many senses,
but giving up on it academically was going too far!?

Hope some guru or mahaguru can enlighten us, thanks!

[r***@gmail.com]

I would like to paraphrase my previous comment:

If you wanted to create an application whose scalability was limited
only by your budget, which platforms/tools would you choose? I
understand there are domain specific answers, eg the quest for
Mersenne primes is served wonderfully by some old-tech C code and a
central repository of results, I guess I am looking for a general or
visionary answer.

[russell kym horsell]

***@gmail.com <***@gmail.com> wrote:
} I would like to paraphrase my previous comment:

} If you wanted to create an application whose scalability was limited
} only by your budget, which platforms/tools would you choose? I
} understand there are domain specific answers, eg the quest for
} Mersenne primes is served wonderfully by some old-tech C code and a
} central repository of results, I guess I am looking for a general or
} visionary answer.

One decomposes problem into relevant modules/units and finds unit(s) with
maximum bang per buck.
Then scale up.

[Jacob Sparre Andersen]

Post by r***@gmail.com
I was looking around for languages for parallel and distributed
programming, [...]

Have you looked at Ada?

Greetings,

Jacob

[Eugene Miya]

Post by r***@gmail.com
I was looking around for languages for parallel and distributed
programming, and I have mostly discovered ruins.

You have only discovered ruins, and you will so far only find ruins.

Post by r***@gmail.com
Eg There used to be
something called CxC and the company changed direction, there were
some systems like Concert, Code http://www.cs.utexas.edu/users/code/
and many others I didnt bother jotting down than have seen no active
development in years or decades. Correct me if I am wrong, it looks
like all the effort is in libraries like MPI and Jini, which is not
the same thing. I think the only project which shows some vitality is
Cilk. Other dead bodies recovered include some distributed Java JVMs
and distributed Python. The most interesting alternative and antidote
to this trend seems to me to be http://boinc.berkeley.edu/

MPI really isn't a language (it is a library and it is widely used).
In the end only time will really tell.

Post by r***@gmail.com
Have I missed something big? Is today's scalability and "speedability"
option simply the conversion and distribution of tasks into batch jobs
(as in Sun's grid) or web service calls (as in Amazon''s cloud)?. I am
guessing distributed programming proved to be too hard in many senses,
but giving up on it academically was going too far!?

If something were big, you would have seen it. Using your MPI example,
there'd be a news group as well as a web page, a wiki, etc.
It's all quite hard still, and quite a few people don't even wants to
get into MPI.

Post by r***@gmail.com
Hope some guru or mahaguru can enlighten us, thanks!

No gurus here. No silver bullets.

There's conferences and journals devoted to solely parallel languages
and most of the other non-language oriented parallelism conferences will
typically have a language/compiler session or there will be a parallel
processing session in languages conferences.

--

[Victor Eijkhout]

Post by r***@gmail.com
I think the only project which shows some vitality is
Cilk.

That one is still on my todo list.

Co-Array Fortran: not a lot easier to program than MPI, and at best the
same performance.

UPC: very easy to program. Can be inefficient if you don't make things
easy for the system.

Titanium: dunno yet. I'm guessing like UPC but don't hold me to that.

Chapel: sounds very interesting, but doesn't exist yet.

Victor.

[Beliavsky]

Post by Victor Eijkhout

Post by r***@gmail.com
I think the only project which shows some vitality is
Cilk.

That one is still on my todo list.
Co-ArrayFortran: not a lot easier to program than MPI, and at best the
same performance.

I can't comment on performance, but noting the OP's observation about
moribund parallel programming languages, let me point out that Co-
Array Fortran will be part of the Fortran 2008 standard.
Representatives of the major hardware serve on the standards
committee, so I'd guess that CAF will become widely available in the
future.