Re: Isn't this in favour of Ada??

[Adrien Plisson <aplisson-news@stochastique.net>]

Tassilo v. Parseval wrote:
> I don't think that Ada tasks and MPI are comparable. MPI is mainly used
> in the fields of supercomputing. You would certainly not use it as a
> fork or threads replacement (incidentally, the 1.x standard of MPI did
> not allow to create processes dynamically) because it would just be too
> painful.

the Ada tasking model may be regarded as a form of MPI, when MPI is 
taken in its broader form of "message passing interface".

but MPI taken in its narrower form of "MPI standard 
<http://www-unix.mcs.anl.gov/mpi/>" is more closely related to Ada 
Distributed Annex, with which it shares a lot of features, including 
all those cited below:

> On the other hand it offers things not present in Ada tasks. MPI has its
> own datatypes that allow you to specify in detail which parts of a
> data-structure to send to another process, where the appropriate
> conversions are done by MPI's pack routines. If you have a matrix with
> rank 100,000 you don't want to share it all and in Ada you'd explicitely
> have to take precautions against that.
> 
> Furthermore, MPI allows you to write a program that does a calculation
> in parallel on your computer and, say, NEC's earth simulator because
> data-exchange between processors can happen over TCP/IP. So MPI really
> is a way to distribute one application across a huge distributed system
> involving workstations that can be anywhere on the internet.

all this should be posible with the Ada distributed annex (annex E of 
the Ada 95 Reference Manual).

a free implementation of this annex for the GNAT compiler is available 
under the name "Glade" (which is not to be confused with the Gtk+ GUI 
builder).

> Also, there are MPI bindings for Ada which wouldn't be the case if those
> two had the same niches in mind.

strange enough when one knows about annex E. i think it exists for 
inter-operability between softwares written using different languages. 
(Annex E is kinda Ada'ish).

(side note: shouldn't PolyORB provide a personality which would allow 
an Ada Annex E enabled application to inter-operate with a MPI enabled 
amplication ?)

-- 
rien