A selection rule for parallel branch and bound algorithms

Branch and bound algorithms are general purpose intelligent enumeration techniques for solving combinatorial optimization problems. It is considered to be well suited to parallel processing. Typically, an application of parallel processing cannot enlarge the solvable size of combinatorial optimization problems. However, parallel branch and bound algorithms can achieve super-linear speedup versus increasing processing elements. In this case, a problem of larger size can be solved in a practical amount of computing time. In this paper, the authors propose the hybrid selection rule for parallel branch and bound algorithms. Typical selection rules for sequential branch and bound algorithms can be naturally enhanced to parallel one. Using six networked workstations, experimental effectiveness comparisons among several selection rules are presented. The results show that the hybrid selection rule leads to super-linear speedup. [In Japanese.]