Lightweight Kubernetes Pushes Orchestrator to the Edge
Kubernetes, the evolving cluster orchestrator, has gone on a diet, stepping off the scales as a lightweight, resilient clustering tool that switches to autopilot once three or more nodes are clustered.
The slimmed-down version dubbed MicroK8s automatically migrates stored data between nodes to maintain a “quorum” in the event of a production failure, Canonical said this week in unveiling the micro-version of Kubernetes. The Ubuntu OS publisher aims MicroK8 at production workloads increasingly running in cloud and server deployments.
Given the complexities of deploying Kubernetes in production, Canonical is stressing its lightweight version as a “zero-ops” alternative for maintaining cloud-based microservices and micro datacenters used for edge computing applications.
MicroK8s is also promoted as a platform that data scientists can use to build and deploy machine learning model pipelines for production deployments using Kubeflow, a tool for tapping into cloud-native Kubernetes resources.
The “high availability” clustering tool can also withstand the loss of a node that would otherwise halt production workloads, Canonical said. A failsafe serves as a datastore embedded in Kubernetes. Dubbed Dqlite, the storage mechanism reduces memory footprint and automates datastore maintenance. The result is “automatic, autonomous high availability,” Canonical claims.
MicroK8s can also be configured to use etcd, the open source distributed key-value store, the London-based software vendor added.
Optimum nodes are selected automatically to deliver the datastore. If datastore node fails, the next best node replaces it. A control plane keeps API services up and running.
Canonical is promoting MicroK8s as a way to increasing resilience in Kubernetes clusters running on edge nodes such as 5G cell towers, vehicles and servers running in remote offices. The company’s “zero ops” approach addresses the expense of maintaining distributed and preferably unattended micro datacenters and clouds. Configuration changes and security patches are transmitted via “compressed over-the-air updates,” the company said.
Along with autonomous operation, the lightweight implementation of Kubernetes is touted as hardening industrial Internet of Things applications. As industrial IoT workloads such as AI inference shift to the cloud, Canonical is pitching MicroK8s as a way of supporting those cloud native applications on the automated factory floor.
It notes that microservices connectivity based on Apache Kafka and other architectures are a “natural fit” for mission-critical industrial control systems.
The company said it is working with the leading public cloud vendors to enable multi-cloud graphics processors. The GPU acceleration feature includes automatic detection of Nvidia’s Cuda application development tools. Along with defining AI pipelines with Kubeflow, the accelerator passes GPUs to Docker applications for deep learning workloads.