OPTIMIZATION OF CLOUD SCHEDULING USING Q-LEARNING HYPERHEURISTIC ALGORITHM

Abstract

  The challenge of producing an optimal task schedule in cloud environments holds significant importance for both the research community and industry. This scheduling problem is computationally complex and is generally classified as NP-complete, and in many cases, NP-hard. Heuristic and metaheuristic algorithms are used to solve optimization problems. Heuristic algorithms use the process of trial and error to find the answer. Metaheuristic algorithms, on the other hand, identify the answer at a higher level. Therefore, we have proposed a better solution to this problem using Reinforcement Learning based Q-learning algorithm which is extended to act as a hyper-heuristic controlling and helping various evolutionary algorithms which are acting as meta-heuristics. It is hypothesized that this strategy will provide us with a near optimal schedule for a real cloud environment. That is, concretely we propose a hyper-heuristic based on Q-learning which drives various evolutionary algorithms HSA, CS, BAT algorithm and as a result the system outputs a schedule which requires as minimum time as practically possible when executed on a real cloud. We test the proposed algorithm on a real world system Hadoop directly for reliable results. For comparisons we use the state of art scheduling algorithms (deterministic, hyper-heuristic and meta-heuristics) and our results demonstrates a significant performance gain over 80% of comparisons.