Performance and energy optimization on terasort algorithm by task self-resizing

In applications of MapReduce, Terasort is one of the most successful ones, which has helped Hadoop to win the Sort Benchmark three times. While Terasort is known for its sorting speed on big data, its performance and energy consumption still can be optimized. We have analyzed the characteristics of Terasort and have identified the existence of idle notes, which does not only waste energy but also loses performance. Therefore, we optimize Terasort through a single-task distributed algorithm and a task self-resizing algorithm to save time and reduce the energy that is consumed by map nodes, which is caused by waiting for tasks and reduce nodes waiting for input. The algorithm proposed in this paper has proved to be effective in optimizing performance and energy consumption through a series of experiments. It can also be adapted to other applications in the MapReduce environment