HyperSpark: A Data-Intensive Programming Environment for Parallel Metaheuristics

Metaheuristics are search procedures used to solve complex, often intractable problems for which other approaches are unsuitable or unable to provide solutions in reasonable times. Although computing power has grown exponentially with the onset of Cloud Computing and Big Data platforms, the domain of metaheuristics has not yet taken full advantage of this new potential. In this paper, we address this gap by proposing HyperSpark, an optimization framework for the scalable execution of user-defined, computationally-intensive heuristics. We designed HyperSpark as a flexible tool meant to harness the benefits (e.g., scalability by design) and features (e.g., a simple programming model or ad-hoc infrastructure tuning) of state-of-the-art big data technology for the benefit of optimization methods. We elaborate on HyperSpark and assess its validity and generality on a library implementing several metaheuristics for the Permutation Flow-Shop Problem (PFSP). We observe that HyperSpark results are comparable with the best tools and solutions from the literature. We conclude that our proof-of-concept shows great potential for further research and practical use.