Summary: | 碩士 === 國立清華大學 === 資訊科學研究所 === 85 ===
Integer Programming Problems are numerical optimization problems that arise frequently in operations research. Their classical solution method is a tree-search based branch and bound algorithm in which each tree node represents a linear programming problem. Although branch and bound is a widespread technique for searching an optimal solution, the running time is still not polynomially bounded. Consequently, the use of parallelism to speed up the execution of branch and bound algorithm has attracted many researchers in recent years. In the previous works, the implementation architectures used are multiprocessor systems, such as n CUBE or CM-5, that the processors are connected by high speed interconnection networks. Performance of that kind of systems are high, though, they are very expensive as for solving integer linear programming problems.
In this thesis, an efficient parallel branch and bound algorithm is proposed. The implementation of our parallel branch and bound algorithm is carried out on a distributed computing software environment called parallel Virtual Machine (PVM). In the PVM environment, several workstations are linked through networks to from a single computing resource.
To demonstrate the performance of our parallel solution, we solved many test problems from the real-world mixed integer programming problem test set MIPLIB. Each problem has been run in the combinations of 1, 2, 4, 8, 16 and 32 processors. Moreover, the test environment consists of 32 IBM PS2 workstations which are connected through 10M bps ethernet.
|