Performance Evaluation of Cloud Task Scheduling Algorithms Using CloudSim Simulation

Abstract

Cloud computing provides scalable, flexible and on-demand access to computing resources through the Internet. As the number of cloud users and computational workloads increases, task scheduling becomes a critical process for assigning cloudlets to available virtual machines and maintaining efficient resource use. Ineffective scheduling can increase execution delay, reduce throughput and leave virtual machine resources underutilised. This study evaluates the performance of cloud task scheduling algorithms using the CloudSim Plus simulation framework. The simulated environment comprised one datacentre with three hosts, eight heterogeneous virtual machines and twenty-five cloudlets. Seven algorithms were implemented and compared: First Come First Served, Round Robin, Shortest Job First, Priority Scheduling, Min-Min, Max-Min and Genetic Algorithm-based scheduling. Their performance was assessed using makespan, throughput, execution time, turnaround time, waiting time, speedup and resource utilisation. The simulation results indicate that Max-Min achieved the strongest overall performance among the evaluated algorithms. It recorded the lowest makespan of 16.40 seconds, the highest throughput of 0.4879 tasks/s, the highest resource utilisation of 66.09% and the highest speedup of 5.2870. Genetic Algorithm-based scheduling also showed favourable performance, particularly in workload balancing and resource utilisation. In contrast, First Come First Served, Round Robin and Shortest Job First produced similar moderate results, while Priority Scheduling performed the weakest under the defined simulation conditions. The findings suggest that heuristic and optimisation-based scheduling approaches can improve cloudlet allocation and resource utilisation in heterogeneous simulated cloud environments. However, the results should be interpreted within the limitations of simulation-based evaluation, as real cloud deployments may involve dynamic workloads, network delays, virtualisation overhead and operational constraints not fully represented in the model.