The 100 Prisoners Problem: Parallel Execution Using Graphics Processing Unit

Abstract

An existing optimal strategy to solve the 100 prisoners problem is to assume that its success probability is independent of the number of prisoners. However, the execution time depends on the size of the problem. For this strategy, both sequential and parallel implementations are applicable. In this paper, we compared the execution times of the sequential and parallel algorithms to see how they vary when the problem size increases. This paper posits that in spite of the parallel nature of this strategy, it will not fully benefit from the GPU implementation. The results show that in spite of the GPU's high memory latency overhead, the parallel implementation will outperform the sequential of larger problem sizes. For the problem size of 100, the GPU implementation using global memory yields a speedup of 0.012. The achieved speedup reaches 1.652, as the problem size increases to 100,000. For the problem size of 100, the implementation using GPU's shared memory runs 8 times faster than the one using global memory.