Using stochastic approximation to design OSPF routing areas that satisfy multiple and diverse end-to-end performance requirements

Dividing an open shortest path first (OSPF) autonomous system (AS) into independent routing areas allows area topology abstraction, reducing route overhead, table size, and convergence time, while providing some isolation from bad routing data. On the contrary, areas reduce connectivity, while increasing configuration complexity, routing path length, and traffic concentration. The formation of performance efficient OSPF areas subject to multiple requirements is known to be NP-complete problem; however, some simple heuristics have been used to optimize for particular routing metrics. For example, a min-cut can be used to ensure balanced number of nodes per area. However, no existing tools can optimize for actual end-to- end performance requirements or take into account the characteristics of network topology. This paper describes a fast and flexible optimization tool that automates the design of open shortest path first (OSPF) routing areas to meet heterogeneous end-to-end performance requirements. The tool is based on an enhanced version of simulated annealing (SA) algorithm, which is a general stochastic approximation method capable of handling multiple, diverse and conflicting requirements (multi-objective optimization). The simulated annealing based tool can provide from highly optimized solutions for network planners designing conventional wired OSPF networks with known application flows to scalability and robust solutions in wireless networks using MANET OSPF extensions with changing application needs. This paper formulates the OSPF areas design as a weighted-sum multi-objective optimization of routing metrics to maximize user capacity, while minimizing the increased delay and lost connectivity. For diverse topologies, we show significantly reduced user delay (over 25%) and increased available bandwidth (by over 400%). Further, we show that by simply adjusting the weights, the tool can prioritize the performance requirements and adapt to heterogeneous network e- - nvironments.