Discrete event simulations are of fundmental importance in designing VLSI
systems, computer networks, manufacturing systems along with a plethora
of other systems. As these systems get larger (hundreds of millions of
gates on a chip, multi core computers, the Intenet...) the demand for
memory and compute cycles to perform these simulations has outstripped
the increase in the number of transistors which can be etched on a chip.
Hence the objective of research in parallel simulation is to take
advantage of parallel and distributed computing platforms (e.g. clusters
of computers) in order to speed up these simulations and to cope with
memory demands. Developing synchronization techniques and load balancing
algorithms constitute the heart of the research agenda in this area.
Continuous simulations such as astrophysical and weather simulations
are equally in need of parallel platforms and have been the subject of
much effort
to parallelize them. The "grand challenge" problems for parallel
computing are in fact concerned parallelizing these simulations in order
to accomodate larger models. Hybrid simulation, the combination of
continuous and discrete event (parallel) simulation provides a new and
powerful computational paradigm.
An overview of this area can be found in
Parallel Discrete Event Simulation-Applications.
Developing synchronization techniques and load balancing algorithms
constitute the heart of the research agenda in distributed simulation.
Most of the work which has been undertaken in the lab in recent years
has been oriented towards distributed VLSI simulation. Another, more
recent project focuses on astrophysics simulations.
Research projects
Distributed VLSI Simulation The importance of this area derives from
the fact that simulation accounts for the bulk of the time spent in
digital circuit design. Consequently, the introduction of parallel
techniques to VLSI simulation will have a major impact on the design of
digital circuits. Our group has developed a distributed Verilog
simulation environment environment (DVS) in order to experiment with
synchronization and partitioning mechanisms for parallel logic and
register transfer level simulation. Verilog is a widely used language
for digital circuit design. Our synchronization algorithms fall into
the class of optimistic algorithms, and more specifically Time Warp.
Astrophysics Simulations Continuous simulations such as
astrophysical and weather simulations have been
singled out as "grand challenge" problems for parallel computing
because of
their importance. Hybrid simulation, a combination of continuous and
discrete event paradigms provides a new and powerful approach to these
simulation problems. In this project we are attempting to integrate
these approaches.
The use of reverse computation is a key component of this reseach.
