Although hierarchical architectures are becoming more and more popular
due to their high scalability and performance yet there has been a
lack of synthesis methodologies that are targeted at such
hierarchically clustered systems.
A novel methodology called MRASM is presented
in this paper for the design of hierarchically clustered parallel systems.
MRASM is a multi-levelrule-based synthesis approach to
parallel architecture design. The synthesized architectures
at each level are simulated and checked for constraint satisfaction.
An intelligent design space exploration method is incorporated into the
simulation environment so that the target system is improved upon iteratively.
One of the various alternative architectures is selected
based on factors such as cost, throughput, utilization, bandwidth,
scalability, reliability, and fault-tolerance. Experiments show that our
approach is indeed feasible.