Performance analysis & improvement of SNPHAP on Multi-core CPUs

In this paper, we attempt to analyse this highly computational problem by parallelizing a haplotype inference algorithm, called SNPHAP. The analysis is based on both the orignal (sequential) algorithm and its corresponding run time complexity in Big-O notations. Then, we improve its performance using OpenMP 3.0 and test on a 4-core Intel Core i7-2600 (Hyper-Threading), an 8-core Intel XeonE5405, an 8-core Intel Xeon E5520 (Hyper-Threading) and a 32-core AMD Opteron 8356 Linux machines. The achievements in terms of maximum speedups are 260%, 316%, 410% and 488%, respectively. The factors that affect the speedup of SNPHAP are the specific parallelized code fraction, the suitable OpenMP constructs, the number of physical cores, the sizes of cache memories within/among CPU cores, the clock frequency and finally the memory technology.