Azure Arc is Microsoft's offering for allowing customers to bring Azure services and management to any infrastructure, including AWS and Google Cloud. This year, during the virtual Ignite conference, the company announced the preview of Azure Arc-enabled machine learning, which extends Azure machine learning capabilities to hybrid and multi-cloud environments.
Microsoft launched Azure Arc in November 2019 at their Ignite conference, and the service received support for Kubernetes - announced during the Build conference 2020. Furthermore, the company brought more capabilities to Azure Arc, which they announced at Ignite 2020 with Azure Arc enabled data services. And now, at this year's Ignite, Microsoft continues adding capabilities to the service with Arc-enabled machine learning. This capability allows customers to build models with Azure Machine Learning anywhere, including on-premises, multi-cloud environments, and at the edge. Moreover, customers can use any Kubernetes cluster and extend machine learning to run close to where the data lives.
With a few clicks, according to a Microsoft Tech Community blog post by Saurya Das, customers can enable the Azure Machine Learning agent to run on any OSS Kubernetes cluster that Azure Arc supports:
Along with other key design patterns, it ensures a seamless set up of the agent on any OSS Kubernetes cluster such as AKS, RedHat OpenShift, managed Kubernetes services from other cloud providers.
When an agent is successfully deployed, an IT operator can either grant data scientists access to the entire cluster or a slice of the cluster, using native concepts such as namespaces and node selectors - making the configuration and lifecycle management of the cluster (setting up autoscaling, upgrading to newer Kubernetes versions) transparent and flexible.
By having IT operators manage the clusters, data scientist do have to know much about Kubernetes and thus can focus on models and work with tools such as the Azure Machine Learning Studio, Azure Machine Learning Python SDK (Software Development Kits), or OSS tools (Jupyter notebooks, TensorFlow, PyTorch, etc.)
Microsoft's Azure Arc offering competes with other comparable offerings from cloud providers such as Google and AWS. Google launched Athos into general availability in April 2019 and brought multi-cloud support in 2020, followed by bare-metal support later in December. Anthos is a more Kubernetes-centric solution than Arc, which brings support for containerized workloads and full support for conventional virtual machines. Next to Arc and Anthos, AWS has Outposts, the hybrid cloud offering from Amazon released in GA last year in January, followed by several updates, including the recently added support for local snapshots. Note that the solution from AWS is compatible only with workloads hosted on AWS's own cloud services.
Holger Mueller, principal analyst and vice president at Constellation Research Inc., told InfoQ:
Building code assets that make a difference for enterprises is not easy and comes with a cost. Naturally, CxOs want to protect these investments and run these code assets everywhere they need them - and this is what the latest version of Azure Arc supports for AI / ML workloads, allowing them to move workloads to Arc locations.
Lastly, customers can sign up for the preview available on March 31st.