Agent-Based Intelligent Parallelization for Multi-Core Applications

Processor manufacturers are no longer able to rely on near Moore’s Law like increases in straight line instruction throughput and clock frequency to provide users with the performance increases that they have become accustomed to. Instead, hardware developers have begun to take a different approach to capitalizing on the increasing number of transistors available by developing processors with multiple execution cores. Areté Associates will develop a scalable, agent-based, parallel processing architecture enabling application developers to efficiently utilize multi-core processors in environments including high-end and embedded systems. This object-oriented software architecture will be embodied in a C++ class library that will provide software developers with parallel processing capability while insulating them from details of the processing hardware. The approach used to develop this capability is an extension of an innovative algorithm component framework being developed at Areté for expediting the migration of algorithms to embedded processing systems. This architecture will be extended to provide a framework and tools for developing multi-core processing algorithms through a combination of a multi-agent system coupled with an optimized task mapping algorithm.