AntElements

Abstract

Ant Colony Optimization (ACO) has become a popular metaheuristic approach for solving hard combinatorial optimization problems. However, most existing ACO software systems are domain-specific, dedicated to concrete problems or non-extensible, non-portable and non-scalable solutions that have been evaluated for problem spaces of limited size. In this context, we present AntElements (AntE), a portable Java-based ACO middleware, designed and implemented with the highest consideration for versatility. The extensibility of the proposed middleware allows its use in virtually any ACO deployment, ranging from experimental to commercial. In this work, the overall object-oriented architecture and the software design patterns of AntE are explained, alongside the main concepts behind them. Furthermore, AntE is analyzed with respect to the computational efficiency, parallelization capacities and memory consumption, which allows to establish its usability and scalability range. In its current implementation, on an average to mid-high workstation, our middleware is capable of processing upwards of 10 5 agents per second, in graphs of the order of 10 5 nodes and sustain a stable, fully logged experiment for over 12 hours. The proposed middleware has already been deployed in several research projects that are outlined in this paper, illustrating the range of possibilities it offers.