Constrained Multi-objective Optimization Method for Practical Scientific Workflow Resource Selection

Courtney Powell, Katsunori Miura, Masaharu Munetomo

Abstract: This paper presents and evaluates a constrained multi-objective optimization method for scientific workflow resource selection that uses equivalent transformation for constraint handling. Two different approaches are compared using a case study of optimal cloud resource configuration selection for a practical genomic analytics workflow. In the first approach, called the nondominated sorting equivalent transformation (NSET) method, feasible configurations are generated via equivalent transformation and the Pareto-optimal configurations are selected from among them via a process of nondominated sorting, reference points association, and niching/elitism. In the second approach, Pareto-optimal configurations are generated via the nondominated sorting genetic algorithms II/III (NSGA-II/III) and feasible configurations are generated via equivalent transformation. Then, the configurations that are common to both processes are considered to be both feasible and optimal. Preliminary results based on the Pareto-optimal configuration sets generated by NSGA-II/III indicate that NSET is feasible for constrained multi-objective optimization of practical scientific workflow resource selection problems.