A Time and Cost-based Matching Strategy for Data Parallelizable Tasks of Grid Workflows

Efficient exploitation of grids for running scientific workflows could benefit of resource brokering systems to automatically and transparently allocate tasks to available resources in the Internet granting the fulfillment of functional and QoS constraints. Existing works typically do not deal with business models to map tasks to resources. Since the service oriented approach is fostering a new vision of grid computing, economic aspects will become key factors to burst the adoption of computing as a utility. This paper presents a time and cost-constrained matching strategy that, according to the data parallelism pattern, is able to deploy a scientific workflow task on a pool of resources selected with the aim of minimizing its execution time. The strategy was implemented in a grid broker and its validity was experimentally analyzed with a real grid of clusters and workstations.