Scheduling for Fast Turnaround Time on Institutional Desktop grid

Institutional desktop grids, that is, grids comprised by the desktop machines of an institution (academic or corporate) can act as an important source of computing power for local user, if the high volatility of such resources is properly harnessed. In this paper, we focus on scheduling techniques that can improve turnaround time of bagof-tasks applications. For that purpose, we combine shared-checkpoint management with several scheduling methods like FCFS, adaptive timeout, simple replication and short-term availability predictions. The main goal is to minimize turnaround time by reducing negative effects of resources volatility on the executions of bag-of-tasks applications. We present and assess several scheduling policies oriented toward fast turnaround times, and assess them through trace-based simulations over the resources of 32 machines of two classrooms of an academic