In June, we announced a cloud interoperability partnership between Microsoft and Oracle. This cross-cloud interconnect enables you to migrate and run mission-critical enterprise workloads across Microsoft Azure and Oracle Cloud Infrastructure. By enabling you to run one part of a workload in Azure and another part of the same workload in Oracle Cloud, this partnership delivers a highly optimized, best-of-both-clouds experience.

This dedicated virtual circuit also enables you to deploy Kubernetes clusters that span Oracle Cloud Infrastructure and Microsoft Azure. We’ve created a step-by-step guide for deploying a cross-cloud, GPU-enabled Kubernetes cluster running on Oracle Cloud Infrastructure and Microsoft Azure using the interconnect. Although the guide walks through deploying a GPU-enabled Kubernetes cluster, you can deploy any workload depending on your needs.

Diagram that shows the architecture of a Kubernetes cluster running on Oracle Cloud Infrastructure and Azure through the interconnect.

How Is This Different Than Using Kubernetes Federation (KubeFed)?

The cluster that we create in the guide doesn’t use Kubernetes Federation (KubeFed). In a KubeFed scenario, you would have multiple separate Kubernetes clusters, and then use KubeFed to coordinate the configuration of these multiple clusters from a single API. The KubeFed project is in alpha release.

In our scenario, you deploy a single Kubernetes cluster that spans Oracle Cloud Infrastructure and Azure. Think of the interconnect as an extension of your network that gives you a dedicated virtual circuit between Oracle Cloud Infrastructure and Azure. So, you can deploy a Kubernetes cluster (or any other cluster) by following the regular steps that you would follow when deploying a Kubernetes cluster on a single cloud or on premises. And after deployment, you are managing a single cluster.

How Low Is the Latency Between Oracle Cloud Infrastructure and Microsoft Azure?

The average ping latency of the interconnect with Azure FastPath enabled is about 1.3 ms. You can find a detailed latency test in this blog post. 

Why Do You Need a Cross-Cloud Kubernetes Cluster?

Adopting a multicloud strategy has many benefits. Our specific scenario, deploying a GPU-enabled Kubernetes cluster for high-performance computing (HPC) workloads across Oracle Cloud Infrastructure and Azure, has three main benefits.

Optimized ROI and Resource Utilization

Many of our customers use multiple clouds. One of their pain points is optimizing the utilization of the resources that they consume. Using a cross-cloud cluster enables you to selectively run workloads across Oracle Cloud Infrastructure and Azure depending on your requirements. It’s easy to tell Kubernetes how you want the job to be run.

For example, you can tell Kubernetes “I need this MPI job to run only on Kubernetes nodes that have NVIDIA Tesla V100 GPUs” or “I need to run this job in the nodes that are in our development namespace and located in the UK.”

Security

The interconnect gives you a dedicated virtual circuit between Oracle Cloud Infrastructure and Azure. The traffic doesn’t go through the internet. You can even deploy a private Kubernetes cluster with nodes that have no public IP addresses.

High Availability and Disaster Recovery

Your cluster runs on both Oracle Cloud Infrastructure and Azure. If an issue occurs with one provider, your cluster is still accessible.

If you have any questions, create an issue in the project’s repo.