A generic parallel optimization framework for solving hard problems in optical networks

Many optimization problems in optical networks are either NP-hard or take a long time to solve. In this paper, we develop a generic parallel computing framework which either solves these problems completely or at least improves the quality of the solution. To verify the effectiveness of the proposed framework, we build a miniaturized parallel computing system and test our approach based on different heuristic and meta-heuristic algorithms for the bin-packing problems, mixed integer linear programming (MILP) models, and distributed machine learning. Experimental studies show that the proposed parallel computing framework based on the miniaturized parallel computing system is effective in so far as it either significantly reduces the computing time or improves the quality of the solution as compared to when only a single machine is used.