Energy-Aware Task Allocation for Large Task Sets on Heterogeneous Multiprocessor Systems

In recent decades, the research of energy-aware scheduling on heterogeneous multiprocessor systems is becoming more and more popular. A classic method for real-time task allocation is Linear Programming (LP). However, existing LP formulations are usually regarded as ineffective in solving large-scale allocation problems due to the unacceptable time consumption. In this work, we propose two integer linear programming (ILP) formulations to deal with the allocation problems for large task sets. One exact ILP(1) is formulated to derive an intermediate solution, and the other relaxed ILP(2) is considered to calculate the desired minimum energy. Then the desired minimum energy can be taken as a reference to evaluate the optimality of the intermediate solution. Experimental results on randomly generated task sets demonstrate that our method achieves average 19.2% less energy within limited time than the classic greedy and the state-of-the-art heuristic algorithm.