Grid Framework for Parallel Investigations of Spiking Neural Microcircuits
Ioan Lucian Muntean, Marius Joldos
Computer Science Department
Technical University of Cluj-Napoca
Romania
Email: {Ioan.Lucian.Muntean, Marius.Joldos}@cs.utcluj.ro
Abstract
Simulation of spiking neural networks is computa-
tionally expensive and the employment of multicore
processors can boost the performance of such simu-
lations. Designing parallelization strategies that work
well for different characteristics of the microcircuits
entails expensive computations, leading to increased
development times. To speed up the design of mul-
ticore software for computational neuroscience, we
have developed a framework that exploits multicore
systems available in grid computing environments. Due
to the use of GridSFEA plugins, common operations
such as evaluation of parallelization strategies can be
undertaken with very little effort.
We evaluated the plugins for the development of a
synchronous multicore spiking neural simulator. This
uses the spike response model combined with the phe-
nomenological model of spike time dependent synapse
plasticity. The parallelization uses OpenMP, the micro-
circuits have small world topologies and count up to
10
4
neurons and 10
7
synapses with biological details.
With this novel framework more complex investigations
in computational neuroscience such as analysis of the
dynamics of neural microcircuits could be tackled.
1. Motivation
The computer simulation of biological neural micro-
circuits has become a necessary instrument for modern
computational neuroscientists. The more realistic the
neuronal models and the interconnecting microcircuit
topologies are, the more computationally expensive
their simulation gets. Their rapid evolution, increasing
availability, and potential for parallel processing make
multicore and manycore technologies attractive to the
simulation codes of neuroscientists.
The adoption of these novel parallelization tech-
nologies in the computational neuroscience software
is rather slow, especially in the area of the simulation
of biological neural microcircuits. The main reason
for this reticence is that such neural systems exhibit
a highly dynamic behavior, close to chaos [1]. Their
simulation is very sensitive to any changes of the en-
vironment (initial conditions, numerical accuracy [2],
microcircuit topology [3], order of the neuronal pro-
cessing etc). Thus, the effects introduced by the paral-
lelization techniques and technologies need first to be
investigated.
To speed up the design of multicore software for
computational neuroscience, we proposed in [4] a
framework for the analysis of the dynamic behavior of
biological neural microcircuits simulated with multi-
core and manycore technologies. In [4], the OpenMP-
based simulation of the Hodgkin-Huxley model was
compared with similar scenarios computed with Neu-
ron [5]. In this paper, in order to foster the exploitation
of multicore systems available in distributed comput-
ing environments we extend the framework with grid
computing capabilities. The extensions are based on
the concept of application plugins described in [6].
We used the plugins during the development of a
synchronous multicore simulator based on the Spike
Response Model (SRM) as presented in [7], combined
with the phenomenological model of Spike Time De-
pendent synapse Plasticity (STDP). The parallelization
is carried out with OpenMP, the microcircuits have
small world topologies and count up to 10
4
neurons
and more than 10
7
synapses with biological details.
Thus, the framework takes specifications of the
simulation scenario, starts corresponding grid jobs, and
handles the user interaction with the jobs and with
their results. As such, common operations such as the
evaluation of parallelization strategies, entailing the
computation of thousands of scenario configurations,
can be defined and undertaken with very little manage-
ment effort. With this novel framework, more complex
investigations such as the analysis of the dynamics of
2012 11th International Symposium on Parallel and Distributed Computing
978-0-7695-4805-0/12 $26.00 © 2012 IEEE
DOI 10.1109/ISPDC.2012.37
219