A Structured Representation for Parallel Algorithm Design on Multicomputers
Xian‐He Sun, Lionel Ming-shuan Ni · 2005
A notation has been developed. Using this notation, most of the frequently used scientific and engineering applications can be represented by simple formulas. These formulas constitute the structured representation of the corresponding algorithms. The structured representations are simple, easy to understand, and contain sufficient information about uneven allocation and communication latency degradations. With the structured representations, algorithms can be compared, classified and partitioned. Some of the basic building blocks, called computation models, of frequently used algorithms, then, are identified and studied. Structured representations relate general algorithms to computation models. Studying computation models leads to a guideline for efficient parallel algorithm design.