Abstract-Production grids exhibit high failure rates hampering the development of many large scale scientific applications. End users require robust experiment production environments ensuring efficient resubmission of failed tasks. Proper parameterization of resubmission strategies is a complex problem that depends on the non-stationary workload conditions experienced by the infrastructure. In order to determine optimal resubmission parameters, probabilistic models of the overhead experienced by grid jobs are defined, taking into account the distribution of faults as measured on the infrastructure. Two strategies that can be implemented on the client side are proposed. Their models are evaluated under variable workload conditions to assess their validity along time. Their results are compared and a trade-off between usability and model accuracy is discussed.