Using Cluster for Fixing Kernel Transpose to Improve Parallel of Strength Pareto Evolutionary Algorithm 2 (pSPEA2)

The motif discovery problems such as DNA motifs need effective methods to search and process due to high computational loads. One of the problems needs to be solved is called Multi-Objective Optimization Problems (MOPS). To solve MOPS requires Evolutionary Algorithms, for instances what is called Strength Pareto Evolutionary Algorithm 2 (SPEA2). But SPEA2 is a computational extensive method. To solve this problem, paralleling the method is often conducted to cut the computational time of SPEA2. In our previous work, we used CUDA platform to parallel SPEA2. The method we called Parallel of Strength Pareto Evolutionary Algorithm 2 (pSPEA2) showed an improved performance over SPEA2. However, since the data are not coalesced, they require computations of kernel transpose which require high communications between the GPUs and the CPU. Hence, pSPEA2 is not optimum in CUDA platform. In this paper, we propose applying pSPEA2 in a cluster. We expect, we could parallel the kernel computation, hence an improved processing speed is expected. Our experiments confirm that the proposed method is 4.65 times faster than our previous method.