Scheduling Tasks with Precedence Constraints on Hybrid Multi-core Machines

In this work, we are interested in scheduling dependent tasks for hybrid parallel multi-core machines, composed of CPUs with additional accelerators (GPUs). The objective is to minimize the make span, which is a crucial problem for reaching the potential of new platforms in High Performance Computing. We provide an approximation algorithm with a performance guarantee of 6 to solve this problem. The algorithm is a two-phase solving method: a first phase based on rounding the solution provided by solving a linear programming formulation for the assignment of the tasks to the resources. A second phase uses a classical list algorithm to schedule the tasks according to the assignment phase. The proposed approach is the first generic algorithm with a performance guarantee for scheduling tasks with precedence constraints on hybrid platforms with CPUs and GPUs resources.