AbstractsComputer Science

This research develops an OpenMP solution for a pairwise comparison of DNA sequence data strings using an extended Smith-Waterman algorithm. The CSX600 series ClearSpeed accelerator measured in past work is used to establish a baseline. We show comparable performance between a single CSX600 board and the OpenMP solution in order to predict the scaling of the ClearSpeed accelerator, with a similar scheduling overhead to the OpenMP implementation. We developed two implementations of the Smith-Waterman algorithm based on the extended ClearSpeed Implementation. One was developed using stack memory and the other using heap memory. We show that the while the OpenMP solution is faster at small problem sizes using the stack memory, it is limited in the problem size it can handle. We also show that for larger problem sizes to be handled in OpenMP it comes at a cost of a memory initialization overhead using heap memory and the ClearSpeed solution is faster when comparing to the heap memory implementation. We predict the future ClearSpeed solution of a much larger board would excel, quickly overtaking the OpenMP implementation for larger, more realistic problem sizes because of the distributed memory model it uses.