Partitioning a given computational mesh for parallel processing is a very active research topic in parallel computation technology. A large number of solution algorithms exists for the a priori, static partitioning problem. The most efficient algorithms are implemented in program libraries like Chaco or Metis.

Nevertheless, the partitioning problem is known to be combinatorial hard. There is no algorithm producing definitely the best solution. All attempts implement more or less efficient heuristics which may fail due to the special properties of the partitioning problem.

Additionally, the parallel machine used to solve the computational problem and the solution procedure itself have significant influence on the suitability of a given partitioning. We propose a special software tool for evaluating the partitioning results with respect to the characteristics of a chosen parallel machine and the application of an iterative solver.

We approximate the iteration cycle time by a theoretical model of the hard- and software. The comparison of a number of partitionings is based on these practical performance figures.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £66 +P&P)