Escalabilidade de aplicações Bag-of-Tasks em plataformas distribuídas heterogêneas

Abstract

Bag-of-Tasks (BoT) applications are parallel applications composed of independent (i.e., embarrassingly parallel) tasks, which do not communicate with each other, may depend upon one or more input files, and can be executed in any order. BoT applications are very frequent in several scientific areas, and it is the ideal application class for execution on large distributed computing systems composed of hundreds to many thousands of computational resources. This paper focusses on the scalability of BoT applications running on large heterogeneous distributed computing systems organized as a master-slave platform. The results demonstrate that heterogeneous master-slave platforms can achieve higher scalability than homogeneous platforms for the execution of BoT applications, when the computational power of individual nodes in the homogeneous platform is fixed. However, when individual nodes of the homogeneous platform can scale-up, experiments show that master-slave platforms can achieve near linear speedups.