Dynamic Platform-Independent Meta-Algorithms for Graph-Partitioning

摘要

A dynamic platform-independent solver is developed for use with network and graph algorithms of operations research. This solver allows analysts to solve a large variety of problems without writing code. Algorithms from a library can be integrated into a meta-algorithm which also provides easy monitoring of solution progress. The solver, DORS, is demonstrated by heuristically solving a graph-partitioning problem to minimize the number of nodes adjacent to other segments of the partition. The model arises from a network-upgrade project faced by the Defense Information Systems Agency (DISA), a problem with over 200 nodes and 1400 arcs. Solutions are provided on a 266 MHz Pentium II PC using Windows NT 4.0. Eight variants of the problem are solved involving modification to the objective function, constraints on the size of partition segments, and on the number of those segments. DORS (and the meta- algorithm it implements) appears to find a good solution for one of the two problem formulations for DISA, but has difficulty solving the other. Because the solver allows new algorithms to be easily added to create more powerful meta- algorithms, DORS should provide a good solution approach for both problem formulations given a more versatile library of algorithms.