Quality-Aware Energy-Efficient Scheduling of Moldable-Parallel Streaming Computations on Heterogeneous Multicore CPUs with DVFS

We solve the optimization problem of scheduling stream computations with parallelizable, multi-variant tasks onto a heterogeneous parallel system with DVFS, such that the overall energy consumption is minimized while maintaining a given throughput and a given quality for the results. We also explore the energy-quality tradeoff via a Pareto optimization of the multi-objective optimization problem. We provide a solution based on integer linear programming and a heuristic method, and evaluate these by scheduling synthetic task graphs as well as task graphs from real-world stream processing applications. The heuristic is 20% to 25% less effective than ILP but much faster (by 1 to 2 orders of magnitude). While ILP is slower for larger task graphs, it is better at finding feasible solutions.