Scheduling under redundancy

In the present extended abstract we investigate the impact that the scheduling policy has on the performance of redundancy systems when the usual exponentially distributed i.i.d. copies assumption is relaxed. In particular, we investigate the performance, in terms of the total number of jobs in the system, not only for redundancy-oblivious policies, such as FCFS (First-Come-First-Serve) and ROS (Random- Order-of-Service), but also for redundancy-aware policies of the form Π1-?2, where Π1 discriminates among job classes and Π2 discriminates among jobs of the same class. Examples of first-level policies are LRF (Least-Redundant-First) and MRF (Most-Redundant-First), where under LRF, respectively MRF, within a server jobs with fewer copies, respectively more copies, have priority over jobs with more copies, respectively fewer copies. Second-level policies could be FCFS or ROS.