Superconcurrent Processing - A Dynamic Approach to Heterogeneous Parallelism
NAVAL OCEAN SYSTEMS CENTER SAN DIEGO CA
Pagination or Media Count:
This reprint describes an approach to finding and using an optimal, heterogenous suite of processors to solve supercomputing problems. This technique, called superconcurrency, currently works best when the computational requirements are diverse and significant portions of the code are not tightly-coupled. It is also dependent on new methods of benchmarking and code profiling, as well as eventual use of artificial intelligence techniques for intelligent management of the selected superconcurrent suite. This latter technique, combined with anticipated bandwidth increases, will permit much more closely coupled code portions to be distributed on the heterogeneous suite. This paper also presents theoretical and empirical results to show SIMD architectures are faster than vector architectures for processing long vectors. Implications for future architectures and distributed heterogeneous processing in general are also discussed. Keywords Massive parallel processing, High performance computing.
- Computer Hardware