A Resource Management System For Containerized Applications

With the rise of container technology, represented by Docker, developers can package an application into a standard container image and distribute it to different cloud computing platforms uniformly. To deploy and orchestrate container applications, there are several container resource management systems published. These systems generally provide orchestrating and monitoring functions, and ask for the user to specify the requirement of computing resources when submitting the task. But in the production environment, the dynamic fluctuations of application's workload result in that these systems can not adjust and satisfy the resource needs of container's. As a consequence, the performance objects are easily violated.The proposed resource management system for containerized applications makes several contributions. Firstly, it designs a schedule algorithm that can improve server's resource utilization. Secondly, it implements an interference detection model that can recognize container's running status. Thirdly, it designs a dynamical resource adjustment strategy to ensure container's performance. Compared to conventional container management systems, the proposed resource management system for container applications through analyzing containers' runtime resource consumption and dynamically adjusting computing resources to ensure applications can execute regularly. To be specific, the system assigns every application with a specific resource pool, and all containers included in an application would share resources in the resource pool. Based on the resource limits specified by the user when submitting the task, the system will schedule the task, try to balance the resource utilization of the cluster nodes, and prevent the occurrence of uneven distribution of resources. When executing an application, performance interference detection model will monitor running status and analyze the performance indicator of containers'. These performance indicators will be the basis to dynamically adjust computing resources of containers', and make anomalous container return to regular status.Experimental results show that our resource management system for container applications could improve the performance of clusters, comparing to current mainstream systems. Scheduling policy embodied in the system can increase resource utilization of the computing nodes in the cluster. Performance interference detection model can detect mostly container performance anomalies.