A Parallel Computing Framework for Large-Scale Air Traffic Flow Optimization

Abstract

Optimizing nationwide air traffic flow entails computational difficulty as the traffic is generally modeled as a multicommodity network, which involves a huge number of variables. This paper presents a framework that speeds up the optimization. Nationwide air traffic is modeled using a link transmission model (LTM), to which a dual-decomposition method is applied. The large-scale problem is decomposed into a master problem and a number of independent subproblems, which are easy to solve. As a result, the execution of solving the subproblem is parallelizable. A parallel computing framework is based on multiprocessing technology. The master problem is updated on a server, and a client cluster is deployed to finish the subproblems such that the most computationally intensive part of the optimization can be executed in parallel. The server and the clients communicate via Transmission Control Protocol (TCP)/User Datagram Protocol (UDP). An adaptive job allocation method is developed to balance the workload among each client, resulting in maximized utilization of the computing resources. Experiment results show that, compared with an earlier single process solution, the proposed framework considerably increases computational efficiency. The optimization of a 2-h nationwide traffic problem involving 2326 subproblems takes 6 min using ten Dell workstations. The increased computational workload due to the increased number of subproblems can be mitigated by the extension of computer deployment.