Efficiency-Aware Jobs Allocation for e-Science Environments

The allocation of jobs to a set of heterogeneous resources, as in common e-Science environments, is a key issue for their efficient use. To reach a better coupling and improve execution times of applications, a consolidated though challenging opportunity is the benchmarking of resources. In this direction, we propose to exploit two complementary approaches, the micro and application-driven benchmarks, which allow to grasp the behavior of computational contexts with respect to more realistic workloads and to lead to a more efficient assignment of the submitted jobs to resources. To evaluate the appropriateness of our methodology an experimentation on a test bed was carried-out. A first comparison of benchmarking results highlighted differences in performance amongst the two levels of benchmarks. Moreover the appropriateness of adopting benchmarking information to rank resources was also confirmed by examining different allocation strategies through a queuing network simulator.