Convertendo recursos compartilhados no espaço em recursos intermitentes.

Abstract

Computational grids are turning from promise to reality. Of course, this is not happening for all applications at once. Bag-of-Tasks (BoT) applications (parallel applications composed of independent tasks) are, due to their simplicity, the first class of applications to be executed massively in grids (e.g. SETIhome). BoT applications are especially suitable for running on grids as they can run on intermittent resources (i.e. resources without guarantees of availability or reliability). In this scenario, good performance and reliable results are provided by eager schedulers, who use replication to compensate for poor task-processor associations. However, greedy schedulers are not prepared to use shared resources in space (e.g. parallel supercomputers). This is because using a shared resource in space involves submitting a detailed request to the resource scheduler, specifying the number of processors and the amount of time these processors must be allocated - information that greedy schedulers are not prepared to provide. Since shared resources in space are the most powerful computational resources, using them could improve the execution time of BoT applications. This work proposes an automatic way of elaborating such requests to the shared resource in space in order to convert shared resources in space into intermittent, consequently making them naturally usable by greedy schedulers. Such conversion is based on heuristics that aim to decrease the execution time of the applications and allows greedy schedulers to use these powerful computational resources without modifications.