Workflow-based comparison of two Distributed Computing Infrastructures

Abstract

Porting applications to Distributed Computing Infrastructures (DCIs) is eased by the use of workflow abstractions. Yet, estimating the impact of the execution DCI on application performance is difficult due to the heterogeneity of the resources available, middleware and operation models. This paper describes a workflow-based experimental method to acquire objective performance comparison criterions when dealing with completely different DCIs. Experiments were conducted on the European EGI and the French Grid'5000 infrastructures to highlight raw performance variations and identify their causes. The results obtained also show that it is possible to conduct experiments on a production infrastructure with similar reproducibility as on an experimental platform.