Multidimensional Resource Task Scheduling Based on Particle Swarm Optimization in Edge Computing

The default Kubernetes scheduling algorithm, Balanced Resource Allocation (BRA), has limitations in achieving resource balance. It primarily focuses on the instantaneous utilization of CPU and memory while neglecting the coordinated optimization of disk I/O and network bandwidth, resulting in an imbalance in cluster resource utilization, commonly known as the bottleneck effect. To address this issue, this paper proposes a multi-task scheduling model based on multidimensional resource demands. By analyzing Pod resource requests and the real-time availability of cluster resources, the model evaluates scheduling decisions to enhance overall load balancing. The proposed approach enables the simultaneous scheduling of multiple tasks while iteratively adjusting the inertia weight and learning factors in the discrete particle swarm optimization (DPSO) algorithm. Experimental results with different types of applications show that, compared to the baseline scheduling methods, the proposed strategy can flexibly adapt to the varying demands of different task types and effectively reduce the imbalance in the utilization of multi-dimensional cluster resources. This is particularly evident in I/O-intensive scenarios, where the IPSO algorithm demonstrates superior resource adaptation capability and utilization efficiency.