A priority-aware scheduling framework for heterogeneous workloads in container-based cloud

With the uncertainty of a cloud environment and the diversity of workload requirements increasing the scheduling cost of container-based cloud, especially for load spikes of application access, optimizing the utilization efficiency of cloud resources and quality of service is the focus of container cluster technology in the future. Different from traditional virtual machine-based scheduling, containerized applications of heterogeneous workloads bring higher scheduling complexity with its elastic scaling and multi-replicas operation. To tackle this problem, we propose a priority-aware workloads scheduling algorithm PA-CCWS. Firstly, we implement workload characterization and behavior identification, quantify the analysis results with TOPSIS method, generate the workloads priority and build priority scheduling buffer queue. Meanwhile, the model learning is accelerated by the experience replay mechanism that inserts and updates the priority of historical experience through the real-time feedback of actual container scheduling from DDQN. Then, we describe containerized applications oriented deep reinforcement learning scheduling algorithm which combined with the two kinds of priorities, to optimize scheduling decision. Finally, we evaluate the effectiveness of our algorithm in terms of resource utilization, resource imbalance degree and SLA compliance rate, etc. Compared with meta-heuristic algorithm PSOS, mathematical model-based algorithm KCSS and other excellent deep reinforcement learning based scheduling algorithms such as DeepRM-Plus and RLSched applying in the container-based cloud, PA-CCWS shows better resource utilization efficiency and convergence stability in containerized applications scheduling.