A parallelization of adaptive task partitioning algorithms (Q1106613)

The authors are concerned with an adaptive algorithm where each step involves either the completion or partitioning of a task. They first discuss the historical background and their model for macro-supported high-level parallelization and the performance modeling. They argue that for many serial adaptive procedures, their methodology provides a direct translation into reasonably efficient parallel versions. At the final section, they apply the algorithm to the evaluation of an improper double integral \(\int^{1}_{0}[\int^{1-x}_{0}(x^ 2+y^ 2)^{-0.5}dy]dx\) within a relative accuracy \(10^{-6}\), as a typical example. According to their implementation under Unix on the HEP machines at Argonne Labs, total execution time seems to be saturated when the numbers of processor are 8.