Query re OpenMOSIX for parallelizing NEURON simulations

[tcleland]

Hi all --

Has anybody ever used MOSIX or OpenMOSIX (http://openmosix.sourceforge.net) as a platform for using NEURON on a cluster?

Failing that, does anybody have an educated opinion on whether it would be likely to work, given the following information:

OpenMOSIX is a kernel extension turns a set of GNU/Linux workstations into a cluster, effectively turning the network into a multiprocessor machine. Hence, applications don't have to be written for or even aware of OpenMOSIX. However, single processes won't be shared among MOSIXed processors if they aren't sharable among the processors in a genuine multiprocessor machine, so for example running Firefox won't be spread over multiple processors, but if I ran Firefox and Kgraph then they'd probably be put on different processors.

So my question really is: how much can NEURON modeling benefit from an OpenMOSIX cluster alone (assuming I'm not using NEURON's parallel tools, MPI/PVM, etc.). If I kludged a process that spawned new incarnations of NEURON to process regions of parameter space, I figure that would work, but what if I used NEURON more as if I were on a single machine. If I use praxis/MultiRunFitter, for example, does it spawn separate processes that OpenMOSIX would presumably farm out to different processors in parallel? Or is it retained within a single process that wouldn't transparently utilize multiple processors (whether via OpenMOSIX or on a single machine)?

Thanks a lot for any advice/thoughts/opinions!

Thom

[hines]

However, single processes won't be shared among MOSIXed processors if they aren't sharable among the processors in a genuine multiprocessor machine

So MOSIX will not give an advantage since NEURON requires PVM or MPI
to do parallel simulations natively.

retained within a single process that wouldn't transparently utilize multiple processors

This is the present situation.