Escalability Evaluation of High-Performing Applications in Physical Cluster and Cloud
Abstract: The performance analysis of High Performance Computing (HPC) applications has recently been facilitaded by the use of cloud platforms. The grants offered by infrastructure providers to research projects in universities has also been contributing on this migration. However, the performance of HPC applications depends heavily on the Input/Output support of the platform, mostly the intercommunication network between virtual machines. In this study, wecompared the performance of selected applications of the NPB NAS Parallel Benchmark and real applications for geographic queries processing, on a physical cluster of commodities machines and another similar cluster on the Microsoft Azure cloud. Our tests demonstrated a speed-up up to 2.7 times lower in the cloud for cpu-bound applications. Despite the superior quality of the processors in the cloud, our tests have also shown a run time up to 3.0 times higher in the cloud.
Keywords: .
Citation: Thiago Borges de Oliveira; Ariel Lauber de Paula Silva; Italo T. da Cunha; Paulo Afonso P. Júnior. Avaliação Comparativa de Escalabilidade de Aplicações de Alto Desempenho em Cluster Físico e na Nuvem. In: Anais do 14o. WPerformance, Recife, PE, Brasil, 2015, pp. 1--14.
Copy citation in bibtex format.