Data Science at the Edge: A Turnkey Solution for the Public Sector

Data science is the art of applying algorithms to infer intelligence out of a data set. These efforts are performed using a pipeline similar to a DevSecOps lifecycle. Once complete, the software developers take over, writing an accessibility layer and beginning their work to produce a machine-learned application. Data scientists also perform continuous monitoring and validation of their models to track the real-time results, and ensure that no unexpected changes occur in the real word.

KubeFrame for AI-Edge, the new collaborative solution from Red Hat, HPE and NVIDIA, is a DevOps play of an artificial intelligence and machine learning (AI/ML) inference that boasts enterprise compute, security and management all in one package. This solution combines easily portable hardware, powerful open source software and industry-leading security to make real-time AI-driven edge processing a reality.

Allowing data scientists to publish their model to thousands of clusters globally enables low latency processing to be done close to data sources. Architectural changes to IT infrastructure that enables this operating model introduce edge-specific challenges with the uptime, resiliency and security of an enterprise hardened system. The unique requirements around this type of output is what drives Kubernetes, an open source platform and the industry standard in container orchestration technology, and showcases why it has been widely adopted in the data science community.

Red Hat OpenShift, the leading enterprise hybrid cloud, Kubernetes application platform, runs natively in your public clouds, datacenter or on the edge; giving you that consistent operational and developer experience from any location. The KubeFrame appliance model allows for one comprehensive purchase, which includes the hardware and OpenShift, in a turnkey configuration so you can immediately deploy these systems in the field.

There are three additional technologies available in KubeFrame for AI-Edge, the first being the Open Data Hub (ODH). The Open Data Hub is a reference implementation of many open source machine learning tools on top of OpenShift. ODH is deployed as an operator and can be installed onto an OpenShift cluster to set up quick templates for deployments.

Red Hat Advanced Cluster Management (ACM), the second piece of additional software, originated out of the Red Hat partnership with IBM and is used for managing numerous remote clusters. As you start to move out to the edge, centralized cluster operations are crucial. For example, centrally patching all clusters effortlessly. With a few clicks, not only can you patch the Kubernetes cluster and all the OpenShift services that run on Kubernetes, but also the underlying operating system of all clusters in an environment, and from there, run that thousands of times across all clusters.

The final piece, Red Hat OpenShift Container Storage, provides performance and highly flexible storage options to scientists and developers. Object, block and file storage are all made available to meet all storage requirements. Additionally, Red Hat’s NooBaa object storage gateway allows edge storage to be a seamless extension of your existing object storage footprint.

The HPE Edgeline EL8000 standardized hardware platform is customizable with many configuration options to access more or less RAM, increased CPU and different GPUs. This system enables data scientists to plug in many other pieces on top to build out what is needed at that particular edge site. The one thing that makes the EL8000 stand out is it fits the size, weight, and power requirements needed for deployments at the edge, that's often limited in an edge environment.

Hardware capabilities include:

• Size: 9" by 9" by 17", or roughly the size of an airline carry-on.
• Memory: Single socket Intel Xeon 2nd Generation scalable processor (Cascade Lake).
• Software architecture: Four node system, one as a bastion node and the other as a three node hyperconverged OpenShift cluster.
• Security and Management: Runs the same iLO 5 included in Proliant datacenter servers.

KubeFrame for AI-Edge was developed with a goal of running redundant Kubernetes clusters on a minimal but industry-hardened footprint. By bringing Kubernetes clusters and all the supporting infrastructure down to less than a cubic foot, only one Edgeline chassis is needed in environments unable to have a full rack of equipment. In the future, Red Hat plans to develop single node OpenShift clusters and podman enhancements for processing without multi-node clusters. Enhancements coming to the HPE Edgeline includes upgraded cores and the option to add another scalable processor, refining storage at each one of the individual blade levels (up to 30 terabytes per blade) while enhancing memory capacity.

Your public sector organization’s edge strategy needs to be simplified, consolidated and standardized. Red Hat OpenShift supports the entire data science pipeline, from data collection through application development, for data scientists and developers alike.

KubeFrame for AI-Edge provides the management, security and processing power you need to place accelerators at the edge in our evolving world of AI. When you're branching out to the edge, your focus should not be learning how to install the software, but rather how to deploy it through enterprise management, enterprise security and enterprise compute.