A Profit-Maximum Resource Allocation Approach for Mapreduce in Data Centers

Resource allocation for Mapreduce data processing poses difficult challenges to system administrators in data centers. The extreme scale of Mapreduce applications require an efficiently profitable resource allocation algorithm that minimizes the energy consumption cost while maintaining the highest level of performance. In this paper, we propose a profit-maximum model that minimizes the cost of energy consumption and makespan. By adopting a minimum-weight b-matching rounding algorithm (MBRA) to find an integer solution, then assign map/reduce tasks to individual slots to build a complete resource allocation. Finally, we perform experiments on real workload to evaluate the profit-maximum model and analyze the performance of our proposed algorithm. The results show that MBRA is able to find a near-optimal integer solution that maximizes the profit per unit time in a lower runtime, and it is up to 30%~70% in profit that is better than the current heuristic scheduling algorithm and the rounding algorithm.