Energy-Efficient and High-Performance Processing of Large-Scale Parallel Applications in Data Centers

When a multicore processor in a data center for cloud computing is shared by a large number of parallel tasks of a large-scale parallel application simultaneously, we are facing the problem of allocating the cores to the tasks and schedule the tasks, such that the system performance is optimized or the energy consumption is minimized. The motivation of the present paper is to investigate energy-efficient and high-performance processing of large-scale parallel applications on multicore processors in data centers. In particular, we address scheduling precedence constrained parallel tasks on multicore processors with dynamically variable voltage and speed as combinatorial optimization problems. We point out that our scheduling problems contain four nontrivial subproblems, namely, precedence constraining, system partitioning, task scheduling, and power supplying. We describe our methods to deal with precedence constraints, system partitioning, and task scheduling. We develop our optimal four-level energy/time/power allocation scheme for minimizing schedule length and minimizing energy consumption, analyze the performance of our heuristic algorithms, and derive accurate performance bounds. We demonstrate simulation data, which validate our analytical results.