Impact of the execution context on Grid job performances

In this paper, we examine how the execution context of grid jobs can help to refine submission strategies on a production grid. On this kind of infrastructure, the latency highly impacts performances. We present experiments that quantify the dependencies between the grid latency and both internal and external context parameters on the EGEE infrastructure. We show how job submission managers, job execution sites and the submission date can be statistically correlated to grid performances. 1. Objectives Grids are increasingly used as support infrastructures for different scientific application areas [8, 6, 3]. Several grids, such as the European EGEE grid infrastructure 1 [7] or the Open Science Grid (OSG) 2, have reached a production level quality of service. These large scale and multi-users systems are characterized by a non-stationary load, their heterogeneity and their geographic expansion. As a consequence, non-negligible jobs submission latencies, measured as the time between the jobs submission and the start of their execution, are observed. They are mainly due to queuing systems, network delays and system faults. This system variability highly impacts application performances and thus has to be taken into account [10]. Several initiatives aim at modeling grid infrastructure Workload Management Systems (WMS). Models of the grid latency enable the optimization of job submission parameters such as jobs granularity or the timeout value needed to make the WMS robust against system faults and outliers. Properly modeling a large scale infrastructure is a challenging problem given its heterogeneity and its dynamic behavior. In a previous work, we adopted a probabilistic approach [4] which proved to improve applicatio