[omniORB] omniidl: bounded sequences

[b.keeping@ic.ac.uk]

This is probably evidence that no one actually uses bounded sequences
much - I know I don't! Anyone disagree?

While I'm broadcasting like this, I have another broader issue I'd
like opinions on. We am currently having to decide between
implementing a fairly "traditional" computationally intensive, lowish
bandwidth distributed application (solution of Mixed Integer Linear
Programs) in one of 2 ways -

1 using MPI
2 using CORBA

The starting point is an object-oriented serial code which will
probably be converted to expose its interfaces via CORBA anyway, in
order to break it into components. This is one reason why I am
inclined towards solution 2.

We will probably  want to run on both of two quite different 
target architectures:
1 A network of workstations  (with separate executables on each)
2 A Symmetric Multi-Processor machine (with (probably) a single 
  multithreaded executable)

I know MPI can give good performance on both architectures, but I
suspect using CORBA will reduce the amount of code specific to the
parallel implementation that we have to write, and allow us to retain
the OO design we have already: e.g. with CORBA, object B can build
itself using information from object A by accessing its methods
piecemeal as in the serial case, whereas if we used MPI we would
probably need to "serialise" object A in some way.

So finally, my question (the obvious one!) - does anyone on this list
have relevant experience and opinions on this issue? If so, either
mail me directly or (I suggest) start a new thread.

Regards,
Ben Keeping
Imperial College