Microsoft has announced multiple updates to Azure Kubernetes Service, their managed Kubernetes offering. The announcements include features around AKS virtual nodes, GPU support for Azure Container Instances, and the donation of the Virtual Kubelet project to the Cloud Native Compute Foundation.
Azure Kubernetes Service (AKS) is a fully managed offering that deploys and operates Kubernetes, which also allows end-users to customise node configuration. Like similar services, such as Google Kubernetes Engine and Amazon Elastic Container Service for Kubernetes, Microsoft's target with AKS is to give a simplified experience to build and run Kubernetes container solutions, without the need for in-depth knowledge around the platform. Consequently, according to Daniel Krzyczkowski, Microsoft MVP and senior software developer, there are several benefits from using AKS.
- Kubernetes master and all nodes are deployed and configured
- Azure Kubernetes Service supports the Docker image format
- AKS clusters are created with support for Azure Files and Azure Disks
- Integration with Azure Container Registry (ACR)
- HTTP Application Routing solution makes it easy to access applications deployed to AKS cluster
With the announcement of AKS virtual nodes, it is now possible to use Azure Container Instances (ACI) as host for your AKS containers. Being a serverless platform for running containers, ACI enables the start of containers in seconds, and also provides billing by the second. Therefore, this combination allows an engineer to quickly scale up and down depending on the load on the AKS cluster. AKS is still responsible for the management of the containers, which allows the use of all the tools and capabilities thereof. Getting started with AKS virtual nodes can either go through the portal with the flick of a switch or via the Azure CLI.
Source: https://docs.microsoft.com/en-us/azure/aks/virtual-nodes-portal
While Microsoft recently already announced the support for connecting Azure Container Instance to a virtual network, they have now also announced the support for GPU resources for ACI. By providing support for GPU based resources, this allows for running compute demanding applications, such as CUDA or deep learning. Moreover, using virtual nodes, these GPU resources are also available for use from AKS. To implement these capabilities Microsoft uses NVIDIA Tesla GPUs, initially supporting the K80, P100 and V100 series, while also giving the option to request other GPUs.
In addition, AKS virtual nodes is built on Virtual Kubelet, which Microsoft released as an open-source project allowing to connect Kubernetes to other APIs. In addition to Azure Container Instances, several more providers use Virtual Kubelet to extend their container platform with Kubernetes, as found on the GitHub repository documentation.
- Alibaba Cloud ECI Provider
- Azure Container Instances Provider
- Azure Batch GPU Provider
- AWS Fargate Provider
- Hyper.sh Provider
- Service Fabric Mesh Provider
In the announcement, Microsoft has also declared that they havedonated the Virtual Kubelet project to the Cloud Native Compute Foundation, where it is now part of their sandbox projects, which states this stage has four goals.
- Encourage public visibility of experiments or other early work that can add value to the CNCF mission and build the ingredients of a successful Incubation level project
- Facilitate alignment with existing projects if (and only if) this is desired
- Nurture projects (e.g. via CNCF Service Desk requests)
- Remove possible legal and governance obstacles to adoption and contribution by ensuring all projects adhere to CNCF legal, code of conduct and IP Policy requirements
By donating the project to the CNCF, this allows for even wider adoption of Virtual Kubelet, providing the opportunity to formalize it further.
News Manager update 19th Dec: This article has been corrected to remove a previously incorrect statement "Additionally, by using virtual nodes, it is now also possible to run both Linux and Windows containers on AKS" that was caused by some confusing statements in the source documentation. Please see the discussion in the comments below. Thanks to Jean-Francois Breton for bringing this to our attention,