Every once in a while, I come across a concept that strikes me as a game-changer, and Kubermesh is one. The idea, put into action by the folks at Ocado Technology, uses Google’s Kubernetes, existing hardware and containers to deploy services across a mesh network instead of standing up dedicated servers in a data center.
According to Ocado, Kubermesh could allow developers to remove the data centers, the network and other machines running around a warehouse deployment and leave only the computing nodes and fiber optics to do the work. (Original story)
Like any good open-source project, Kubermesh is available for free download in Github, where it’s described as a “Bare metal, self-hosted, self-healing/provisioning, mesh network kubernetes cluster.”
It works by deploying a single machine and then attaching to it other machines, which can range from low-end Intel NUC boxes to workstations. Using Google Cloud’s Kubernetes container management, Kubermesh adds each node’s resources to the cluster and deploys containers. It’s self-healing and self-provisioning because the orchestration engine knows when a node goes down and can redeploy the affected services to another node.
Such grid computing isn’t new, but it’s growing increasingly popular — and doable — thanks to management tools that dynamically solve networking and resource issues on the fly. In that way, Kubermesh is like Syracuse University’s Orange Grid, which turns humble workstations in offices and computer labs across the Upstate New York campus into a 12,000-core cluster.
According to SU, “this distributed computing system is used by SU faculty and researchers, particularly in the physical sciences and engineering, who need reliable, high throughput computing (HTC).”
Such deployments can make use of existing compute resources, saving time, maintenance and money.