The need to efficiently execute applications in heterogeneous environments is an obstacle parallel computing wants to overcome. The communication heterogeneities present on Multicore clusters ought to be handled carefully for improving efficiency and speedup. For this reason, executions on Multicore environments become a challenge that parallel application programmers must be prepared to deal with. This work proposes an execution methodology centered on controlling communication heterogeneities and improving system efficiency on Multicore clústers. Such methodology is composed by four phases: characterization, a tile distribution model, mapping strategy, and scheduling policy. This method has been developed for Single Program Multiple Data (SPMD) applications with high communication synchronism. We will focus on SPMD applications which are designed through a message-passing library for communication, and selected according to their synchronicity and communications volume. The main contribution of this methodology is to determine the approximate number of cores necessary to achieve a suitable solution in obtaining a good execution time, while the efficiency level is maintained over a threshold defined by users and. Our results show an improvement around 39% in the best case of efficiency on SPMD applications tested when our methodology is applied.
Metodología para la ejecución eficiente de aplicaciones SPMD en clústeres con procesadores multicore
Muresano Caceres, R. R. (Author). 7 Jul 2011
Student thesis: Doctoral thesis
Muresano Caceres, R. R. (Author),
Luque Fadon, E. (Director),
7 Jul 2011Student thesis: Doctoral thesis
Student thesis: Doctoral thesis