Network Bandwidth-Aware Kubernetes Cluster

Currently Kubernetes scheduler only considers cpu and memory but with network bandwidth oversubscription widely exists in the data centers, bandwidth contention is a big problem for bandwidth sensitiv …

Currently Kubernetes scheduler only considers cpu and memory but with network bandwidth oversubscription widely exists in the data centers, bandwidth contention is a big problem for bandwidth sensitive applications. With Kubernetes commonly deployed in a virtualized infrastructure like vSphere or OpenStack, network information at Kubernetes worker node level is not enough. If we want to guarantee those special applications won’t get blocked anywhere in a cluster, we need to reserve network bandwidth for worker nodes in the whole cluster and use that information in the algorithm of scheduler. In this proposal we will not only demonstrate an approach to extend the default Kubernetes scheduler to an external service to make better placement on network resources, but also introduce a method to ensure ongoing network bandwidth for critical applications.