Connectionists: NEURON 5.8 simulates large nets on parallel hardware

[Ted Carnevale]

NEURON version 5.8 is now available from
   http://www.neuron.yale.edu/neuron/install/install.html
This includes many bug fixes for features that were new in the
previous official release of NEURON.

However, for many users the most important changes may be the
improvements that facilitate its use in a parallel computing
environment.  These are:

1.  ParallelContext now works on top of MPI as well as PVM
(see "Top 10 reasons to prefer MPI over PVM"
http://www.lam-mpi.org/mpi/mpi_top10.php).  This makes it much
easier for users to take advantage of NEURON on parallel hardware
that ranges from workstation clusters to massively parallel
supercomputers.  This is potentially helpful for all users who
are faced with optimization, parameter space exploration, and
"large model, long runtime" problems.

2.  NEURON 5.8 also introduces something for users with very large
network models that just won't fit on a single-CPU computer:
the ParallelNetworkManager, which manages an enhanced ParallelContext
to support neural network simulations on parallel machines.  That is,
the ParallelNetworkManager lets you run simulations in which each
processor--be it a node in a supercomputer, or a single- or multiple-
CPU workstation in a cluster--handles a different part of the network.
Tests on large scale problems indicate a nearly linear speedup with
increasing number N of processors, until N is so large that the CPUs
don't have enough work to keep themselves busy.

In making this announcement of NEURON's new capabilities for parallel
computation, I would like to acknowledge the special contributions of
Michele Migliore and Bill Lytton, who collaborated with me on the
design of these features, and Henry Markram, with whom I have been
collaborating on large scale simulation problems and whose IBM Blue
Gene supercomputer has been a most helpful testbed for these new
features.

Michael Hines