An improved beluga whale optimization using ring topology for solving multi-objective task scheduling in cloud

•Proposed a Multi-objective version of Beluga Whale Optimization.•Integrate Opposition-Based Learning into Levy Fight Distribution for searchability.•Improve local search in multi-objective BWO by redesigning ring topology.•Introduce Hybrid Balance factor to provide a good exploration and exploitation.•Verify performance of proposed algorithm against recent and similar techniques. To enhance cloud system performance and customer satisfaction levels, task scheduling must be addressed in the system. Beluga Whale Optimization (BWO) is a metaheuristic method that was developed recently. However, this method still suffers from local minima stagnation despite having an operator that enhances the diversity of population. As a result, Opposition-Based Learning (OBL) can be combined with a Levy Fight Distribution (LFD) and a hybrid balance factor to overcome conventional BWO’s main weaknesses, including slow convergence and local optima traps. We present a multi-objective form of improved BWO (IBWO) to solve task scheduling problems considering both makespan and costs. Multi Objective Improved Beluga Whale Optimization with Ring Topology (MO-IBWO-Ring) is proposed as an efficient task scheduling algorithm that uses whales as feasible solutions for cloud computing tasks. Local search capabilities are also enhanced by using the ring topology concept. The proposed MO-IBWO-Ring algorithm as an optimization algorithm is tested on ten new test functions, and its performance is compared with four algorithms (i.e., Decision space-based Niching Non-dominated Sorting Genetic Algorithm II (DN-NSGAII), Multi-Objective Particle Swarm Optimization algorithm with Ring topology and Special Crowding Distance (MO_Ring_PSO_SCD), Omni-optimizer, and Multi-Objective Particle Swarm Optimization (MOPSO)). Two scenarios have been used to evaluate MO-IBWO-Ring’s performance as a task scheduler. 1) Heterogeneous Computing Scheduling Problem (HCSP) is used as the benchmark dataset with a small (512) and a medium (1024) number of tasks, and 2) with random generated tasks and VMs. When measuring provider metrics, the proposed method achieved better results than competing methods.