Session
Look who's talking!
Connecting to devices at the edge with minimal footprint using AKS Edge Essentials and Akri on WASMs
Yu Jin Kim - Microsoft / Francisco Cabrera - Microsoft
Kubernetes on the edge is becoming increasingly popular for orchestrating workloads closer to where the data is located to reduce costs and latency. However, the edge usually consists of IoT leaf devices that are too small, too old, or too locked down to run Kubernetes on their own (i.e. IP cameras, thermometers, humidity sensors). Contrary to cloud Kubernetes deployments, being at the edge also means having to deal with more constrained resources for your workloads running on clusters deployed on hardware. While containers are great for running workloads, they have high startup costs, bigger binaries and higher memory consumption. So how can these leaf devices be dynamically bridged to a cluster while being as efficient as possible with the available resources at the edge? This talk will go over how users can use AKS Edge Essentials with WASM workloads to deploy a cluster on “light” edge hardware (PC-class devices) and run Akri (a CNCF Sandbox project for exposing leaf devices as resources in a cluster), minimizing the footprint when connecting to these devices and utilizing them in your applications. The talk will include a demo where Akri-connected devices can be utilized in Slight (SpiderLightning) apps through a gRPC proxy running on an AKS Edge Essentials cluster.
Benefits to the Ecosystem:
Modern solutions now span cloud to edge, but Kubernetes at the edge still has its challenges. The main challenge is that it is extremely resource-constrained. Leveraging existing hardware and running workloads with minimal footprint is crucial at the edge. Using WebAssembly significantly reduces resource consumption and increases the responsiveness of the cluster even on older, smaller hardware. AKS Edge Essentials enables deploying K3s or K8s clusters on devices that have memory as little as 4GB and compute as little as 2vCPU. With this resource-constrained cluster, the demo will showcase how to connect to leaf devices with extremely small footprint; by running Akri on a WASM instead of a container, we can make the OCI image 54 times smaller. As a result, the Akri devices can be utilized in Slight apps, consuming far less resources than if they were to be run on traditional containers.