Abstract
SMP Clusters with fat nodes offer an interesting capability for large applications that employ a hybrid parallelization model: to improve load balance, the number of threads can be increased in order to speed-up busy MPI processes or decreased to slow down idle MPI processes, provided these processes reside on the same SMP node. We developed a library which performs this thread adjustment automatically during program execution. Experimental results demonstrate remarkable speed-ups with minimal programming effort. © Springer-Verlag Berlin Heidelberg 2006.
Cite
CITATION STYLE
Spiegel, A., An Mey, D., & Bischof, C. (2006). Hybrid parallelization of CFD applications with dynamic thread balancing. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 3732 LNCS, pp. 433–441). https://doi.org/10.1007/11558958_51
Register to see more suggestions
Mendeley helps you to discover research relevant for your work.
