Kubernetes at its core is a container orchestration system. But simply running containers for their own sake has little purpose, as at the end of the day what really matters are applications.
Among the most interesting and often challenging types of application workloads are machine learning ones, which can often be difficult to deploy and operate. On Dec. 21, the Kubeflow project was officially announced by Google engineers as a new stack to easily deploy and run machine learning workloads.
"The Kubeflow project is dedicated to making Machine Learning on Kubernetes easy, portable and scalable," the Kubeflow GitHub project page states. "Our goal is not to recreate other services, but to provide a straightforward way for spinning up best of breed OSS solutions."
Kubeflow isn't just for any type of machine learning stack, which isn't a surprise given that its development has been led by Google. Rather, Kubeflow uses the Google open-source Tensorflow project as its base.
Kubeflow provides a Tensoflor serving container as well as a Customer Resource (CRD) that enables administrators to easily configure cluster size as well as CPU and GPU resource constraints.
A core tool in many machine learning workloads is Jupyter notebook, which are also part of the Kubeflow.
In addition to a running Kubernetes cluster, Kubefolow also requires that users are running ksonnet, which is an open-source project led by Heptio. Ksonnet was first launched in May and provides a way to easily manage the configuration syntax used in Kubernetes.
"We think working with multiple environments (dev, test, prod) will be the norm for most Kubeflow users," Google engineers wrote in a blog post announcing Kubeflow. "By making environments a first class concept, ksonnet makes it easy for Kubeflow users to easily move their workloads between their different environments."
In an effort to make it easy for users to try out Kubeflow, there is an online Katacoda environment to demo the technology.
Sean Michael Kerner is a senior editor at ServerWatch and InternetNews.com. Follow him on Twitter @TechJournalist.