Vincent Rodin, Gireg Desmeulles, Pascal Ballet, Pascal Redou
and Christophe Le Gal.
An efficient algorithm based on weak synchronization for distributed
in virtuo biological experiments.
Multiagent and Grid Systems, IOS Press, 7(4-5):159-182, November 2011.
Abstract:
Virtual Reality is becoming increasingly necessary to study complex systems
such as biological systems. Thanks to Virtual Reality, the user is placed at
the heart of biological simulations and can carry out experiments as if he
were under the same experimental conditions as in vivo or in vitro.
We usually call this kind of experiments in virtuo experiments.
In order to rapidly develop Virtual Reality applications related to biology,
we have already proposed the RéISCOP meta model which makes it possible
to easily design biological simulations and undertake in virtuo experiments.
This meta model allows to describe a biological system as a composition of its
sub-systems and the interactions between the constituents of these sub-systems.
Unfortunately, when using a single computer, the number of simulated entities
is far from what is needed in biological simulations. It seemed thus necessary
to extend the RéISCOP meta model so that it allows distributed computing
on a grid. We made this choice because the structure of the RéISCOP
meta model is well adapted to a distribution on a grid where the sub-systems
which compose a system can be dispatched on different nodes, the
synchronization and the coherence of the system being ensured by a
Peer-to-Peer architecture.
Unlike traditional approaches which propose a spatial distribution, the method
we describe in this paper is based on an "organizational" distribution linked
to the RéISCOP meta model. This "organizational" distribution is mainly
ensured by using two efficient algorithms based on a dead reckoning method,
one for a data consistency between nodes and one for a weak synchronization of
the nodes involved. These two algorithms are integrated into the behaviors of
agents (DIVAs) which are located on each node of the grid. These agents are
able to communicate by using a Peer-to-Peer architecture upon the grid.
In order to validate our approach, we implement three distributed simulations
with increasing complexities and we compare the results with the results
obtained in the non-distributed simulations. We get very similar results for
the distributed and the non-distributed simulations.
Keywords:
Distributed simulation, Multi-Agent system, Multi-Interaction system,
Peer-to-peer, Virtual reality.
[doi:10.3233/MGS-2011-0173]
[Rodin11a.pdf]