Nodeless OpenShift
OpenShift Container Platform is the leading enterprise Kubernetes platform to build and deliver better applications faster in a cloud agnostic way. With increased availability of varied flavors of cloud compute capacity at varied price points (on-demand, spot, CaaS), Nodeless is emerging as a cloud-native, simple, secure, cost-effective way to consume ephemeral cloud compute for Kubernetes workloads. This blogpost illustrates how Nodeless works with OpenShift Container Platform.
Architecture
With Nodeless OpenShift, Control Plane sees a virtual worker node that advertises large virtual capacity. When Control Plane schedules a pod to the virtual worker node, the worker node provisions just-in-time, right-sized, cost-effective compute on your cloud provider of choice. This compute launch type will be sourced from available launch types on your cloud provider: {on-demand, spot, fargate} on AWS, {dedicated, low-priority, ACI} on Azure, and {normal, preemptible, cloud run} on GCP. The pod is then dispatched to the just-in-time provisioned compute cell. After the pod terminates, underlying compute cell is automatically terminated.
Demo
Advantages
- Pay-as-you-go
Average utilization of Kubernetes clusters on public cloud is 20%. Nodeless prevents wasted resources (wasted spend) by preventing unused worker node resources in your OpenShift cluster.
2. Multi-tenant security
Nodeless delivers separate compute launch type for each pod. This results in stronger (VM-level) isolation between your pods.
3. Operational simplicity
Nodeless enables treating your worker nodes as cattle rather than pets. IT Operators no longer need to hand-curate and babysit worker nodes.
Usecases
- Cloud bursting
Your Control Plane and virtual worker node can sit on cloud provider A (ex: on-premise datacenter), and consume burst capacity from cloud provider B (ex: AWS).
2. CICD
CICD workloads are short-lived and bursty. Hence Nodeless OpenShift is a great fit for Jenkins and other CICD environments.
3. Machine Learning / AI
ML infra is expensive, and ML workloads have batch properties. Use Nodeless OpenShift to prevent wasted spend on expensive cloud GPU devices for your ML workloads.
4. DR
Nodeless OpenShift commoditizes cloud compute. It enables you to easily set up DR environment across multiple cloud vendors and execute fast failover by eliminating the need for hand curating vendor-specific compute for your OpenShift cluster.
Resources
- OpenShift: https://www.openshift.com/learn/get-started/
- Nodeless blogposts: https://medium.com/elotl-blog
- Try Nodeless: https://github.com/elotl/kubeadm-aws
