distributed computing implementations

[Andrew Dalke]

robin:
> If you want the biggest, boldest approach and don't care about
> overhead, use CORBA.

What overhead would this be?  From what I see of omniORB,
there isn't really that much.  Also, CORBA is the most complete,
eg, it allows callbacks and passing around object references
the others don't have.

I've wanted to do something in CORBA for years.  I've
never gotten there.  One problem is that I'm used to Python,
where I don't need to describe the interface beforhand.
CORBA wants that IDL, and a change in the object's interface
 must be reflected in the IDL.  That just seems tedious to
me now.

I also do nearly everything in Python, so don't need the
ability for different langauges to interoperate.  I just pass
around Python objects.

I keep hoping that one of the component systems (like
in GNOME or KDE) takes off, so that "scripting" a la
COM takes off in unix systems, but I've been hoping
for that for the last 5 years.

> If you want a simpler approach but with performance and
> implementation difficulties, use SOAP or XML-RPC.

I wouldn't put SOAP as simple, and I've had problems with
interoperability between various packages.  If I needed to pass
simple data around, I would use XML-RPC.

> If you want the ultimate in simplicity and are willing to foresake
> multi-language support, use Dopy, Pyro, or Twisted Spread.

One reason I'm looking at Twisted is because it handles other
interfaces as well.  I need to talk to SQL databases, SOAP and
XML-RPC servers, straight HTTP, and spawned off external
processes.

There is also older interfaces, like PVM and MPI from the
high-performance computing world, and Linda, and more,
but I haven't looked into them for years.

                    Andrew
                    dalke at dalkescientific.com