AKS Add-ons Part 3: Device Framework

• Azure Kubernetes Service (AKS) and Kubernetes add-ons support devices ranging from GPUs to IoT devices in substantially different scenarios.
• Extensibility plugins in AKS support GPUs in Azure; however, developers must select from several options, some of which are in preview.
• Azure IoT Operations in AKS can connect to a wide variety of IoT standard interfaces and devices using the open-source Akri extension. 

 

Azure Kubernetes Service (AKS) and Kubernetes were originally designed to support the orchestration (or management) of containerized applications running in clusters, that is, scaled-out solutions running on multiple server nodes (physical computers, or more often, VMs). However, certain compute-bound requirements, such as AI, mandate that AKS applications use specialized hardware—most commonly today, graphics processing unit (GPUs). Additionally, in environments such as manufacturing, AKS and Kubernetes are increasingly used at the edge to support remote operations that must connect to, manage, and control IoT devices, often at scale. These two scenarios differ because typically only one to a maximum of four GPUs are ever connected to an AKS node, and they are considered part of the VM hosting the node; whereas IoT environments can be complex and dynamic with hundreds of devices connected and disconnected in real time. Support for both scenarios is ultimately based on Kubernetes’ device plugin framework, through which devices’ availability and capabilities are made available (“advertised”) to Kubernetes applications.