An intelligent and interpretable rule-based metaheuristic approach to task scheduling in cloud systems

Metaheuristic algorithms can be very successful in solving scheduling problems. However, these methods can be slow for time-critical applications due to the iterative stochastic processes they perform. This problem becomes even more pronounced in dynamic environments such as cloud systems. This paper proposes an interpretable rule-based solution to minimize this problem. Combining metaheuristic task scheduling solutions and machine learning techniques, this method uses a two-phase mechanism, semi-offline and online. In the semi-offline phase, we first archive metaheuristic solutions to previously randomly generated or encountered task-scheduling problems. The basic idea of the proposed method is to reuse these previously obtained successful metaheuristic solution patterns for future similar problems. Finding similar solution patterns from this extensive archive dataset is done by automatically extracting rule sets through machine learning techniques. These interpretable rule sets are used to identify the type of task scheduling problem encountered in the online phase and to find the optimal solution pattern. The performance of this method, which dramatically reduces execution time and enables the use of metaheuristics in time-critical applications, has been tested and proven for various cloud task-scheduling scenarios.