Multi-objective task scheduling based on PSO-Ring and intuitionistic fuzzy set

The task scheduler belongs to the NP-complete problem, so it is very challenging in the cloud environment to develop one with reasonable performance and computation speed. Several studies take into account some important factors of the users or providers when addressing cloud task scheduling. This paper models cloud task scheduling as a Multi-objective Optimization Problem (MOP) that maximizes load balancing and execution times. Based on ring topology, we present a new multi-objective particle swarm optimization approach that utilizes intuitionistic fuzzy set to enhance evenness and diversity. The diversity and spread of the Pareto solution is adjusted using a Balanced Intuitionistic Fuzzy Crowding-Distance (BIF-CD) and Intuitionistic Fuzzy non-dominance sorting (IF-dominance). In order to identify the best compromise solution and in decision space and adjust evenness in objective space, ring topology is employed to identify a larger number of Pareto-optimal solutions. Experimental results for 25 benchmark multi-objective functions demonstrate the superiority of the proposed IF-MO-Ring-PSO over four state-of-the-art algorithms. We compare different quantitative measures to assess the uniformity and quality of Pareto fronts (PFs) found by our compared methods. The performance of the proposed method is evaluated on the benchmark test suites ZDT, DTLZ, CF, mDTLZ, and MMF, using the delta and space metrics and the progression metrics. According to the proposed method, ZDT test suite reduced space and delta metric by 13.36 and 15.11% respectively compared to other methods. The Wilcoxon’s test and two-sample Mann Whitney’s test are used to analyze the performance of proposed method on CF test suit. The analysis shows that the proposed method has better ability to generate quality PF with uniformity on constraint test suits. In comparison to four scheduling algorithms, the proposed scheduler also shows better performance according to load balancing, makespan, and resource utilization based on two datasets (i.e., HCSP and GoCJ).