Demystifying Parallel Class for Optimal Performance