Improved task scheduling and load balancing in fog computing infrastructure using hybrid evolutionary method

Task scheduling and virtual machine (VM) placement are two of the main issues in fog computing infrastructure. These elements are closely related to the crucial issue of fog computing's resource and energy usage. Achieving cost-effective execution and improving resource utilization requires careful consideration of task scheduling and VM deployment. Because the task scheduling problem is NP-hard, scientists are exploring metaheuristic algorithms that are inspired by nature. On the other hand, evolutionary algorithms can perform significantly better when their initialization solutions are well-designed. In this paper, a hybrid evolutionary task scheduling (HETS) approach for reliable fog computing task scheduling and VM placement is presented. By combining a new VM placement method with an enhanced particle swarm optimization (PSO) algorithm and using the ability of direct binary. The direct binary encoding technique in PSO represents particle positions as binary values (0 or 1) and velocity within [0, 1], suited for task scheduling problems. A probabilistic approach is used to update binary positions based on the continuous velocity. This method efficiently applies PSO to binary decision spaces, enhancing its applicability to complex optimization problems. Also, a new method was introduced to enhance VM placement, focusing on memory and processor utilization for balanced resource distribution. This approach allocates resources based on memory, processor consumption, and remaining capacity, categorizing physical machines into groups by available resources. This categorization helps select suitable VMs according to tasks and resource needs. The simulation results show that the HETS, compared to other evolutionary algorithms ACO, GASA, and GA, on average, reduces the energy consumption by 5, 9, and 14% and makespan 4, 6, and 11%, respectively. Additionally, the results show a better degree of load balancing and execution time compared to other approaches.