|
From: | Marcus G. Daniels |
Subject: | Re: [Swarm-Modelling] Floating point arithmetic |
Date: | Fri, 06 May 2005 07:51:23 -0600 |
User-agent: | Mozilla Thunderbird 0.9 (Windows/20041103) |
Russell Standish wrote:
You said that Java has an MPI interface too - perhaps that doesn't work too well?
James Marshall wrote:
http://aspen.ucs.indiana.edu/pss/HPJava/mpiJava.html As I say, I haven't used it. It's a JNI interface to the MPI libs though.
My point is that high level interfaces like RMI often don't come for free. Obviously (laugh), I'm a big believer in JNI and similar foreign function interfaces. I don't see a reason why MPI for Java (like Swarm) couldn't, in principle, work well (especially in a GCJ-like framework), but that's changing the argument.
Homogeneity is nice, but there will always be some new technology that comes along that aims to subsume everything else. Sometimes it is good to jump, othertimes it is not. [And no, in retrospect, I don't think rewriting Swarm in Java would have been a good jump. In fact, had I known it was going to be redundant, I would have spent the time developing something more radical. Currently, the HPCS languages Chapel and Fortress look interesting but I don't work on agent-based modeling infrastructure anymore so it is of course just talk. Repast's work with .NET is also interesting to me.]
[Prev in Thread] | Current Thread | [Next in Thread] |