Many Kubernetes clusters support multiple types of online and offline workloads. Due to the traffic fluctuations of online workloads and the variable start time points of offline computation tasks, resource demands vary between peaks and valleys. For example, many enterprises perform dense computing on weekends and during the middle and end of each month. A large amount of computing power is required in these periods to cope with burst computing resource demands.
A common method for Kubernetes cluster scale-out is to use Autoscaler to start new nodes automatically (about one node every 2 minutes), until a pod is scheduled to run. After all tasks in the pod are executed, the temporary nodes are released automatically. In this method, a pod needs to wait for more than 2 minutes before it is scheduled to run.
The Virtual Node addon enables you to address peak demands for computing resources in a Kubernetes cluster at the minimum O&M cost (without changing the number of nodes).
Virtual nodes of a Kubernetes cluster prevent the shortage of computing power and the waste of idle computing nodes, while also providing indefinite elasticity of computing power to help you easily address the peak and valley demands for computing resources.
For more information about how to deploy virtual nodes in a Kubernetes cluster, see Virtual Node Best Practice 2 of Container Service for Kubernetes: Jobs + Virtual Nodes.