Energy Efficient Allocation and Scheduling for DVFS-enabled Multicore Environments using a Multiobjective Evolutionary Algorithm
Zorana Banković, Pedro López-García · 2015
We present an approach for the automatic solving of the scheduling and allocation problem in multicore environments, as well as assigning optimal voltage and frequency levels to each core, using a multiobjective evolutionary algorithm (EA) where both energy consumption and the makespan are optimised and all deadlines are met. The main advantage of our approach is that we deal with all the aspects of the problem at once, which allows searching the whole solution space. In addition, the algorithm introduces the possibility of task migration, which is a novelty in EA-based approaches. Our results show that proper scheduling and allocation can provide significant energy savings in different scenarios: for our test case, and comparing to the well known YDS algorithm, up to 76% on average in the case of loose deadlines, and up 70% on average in the case of tight deadlines can be saved.