The possibilities of connecting data analytics out to the edge often can lead to greater efficiencies and immediate ROI as businesses streamline their analytics orgs and get more from their data. Companies can push image recognition to the edge by enabling surveillance cameras to run inference in real time. Manufacturers can conduct automated quality assurance in factories from one, centralized location. In other cases, these technological leaps can save lives—like when an MRI machine detects an issue in a matter of hours versus weeks.
Dismantling hurdles
Standing in the way of these exciting transformations are some fundamental data problems that span infrastructure, software, and security. Many edge analytics solutions are slowed as data and insights are handed off from data engineers to data scientists, to ML engineers, and finally to business analysts. Friction develops as more hands get involved in aggregating and transforming data, training and validating models, and pushing models into production. Often, these professionals use their own tools and stacks, which creates bottlenecks when migrating or reformatting data or when refactoring apps and transferring them between operating models.
To circumvent these problems, organizations may apply a variety of tools and bespoke technology stacks in an attempt to connect their analytics more efficiently across locations and business units. Often, these Frankensteined systems create redundant copies of data, lock you into proprietary solutions, become too expensive and difficult to manage, and make the collaboration between Data Ops and ML Ops processes inefficient.
Data has evolved, and enterprises must evolve their analytics strategies to match. Vital information has ventured beyond the datacenter and even outside of the cloud. Most data is now originating at the edge, which gives business opportunities for even faster time to insight. It also makes a data center or cloud centric approach to data analytics untenable.
It’s time to think edge-in.
Edge to cloud
Rather than working around these longstanding obstacles, a future-proofed enterprise will choose a unified solution that’s purpose-built for seamlessly connecting the edge to their datacenters and public clouds. That solution requires the right software, infrastructure, and support to create truly streamlined analytics. Nobody can do it on their own.
Hewlett Packard Enterprise (HPE) is proud of our longstanding partnership with NVIDIA as we work to close the gaps and eliminate the pain points in deriving insight from data. And together, we’ve created the first-of-its kind solution to deliver end to end data distributed analytics from edge to cloud.
This next gen analytics stack is the culmination of a long journey. HPE and NVIDIA have been testing and validating the right combination of software, infrastructure, and services behind the scenes to deliver a cohesive, accelerated edge-to-cloud platform that’s ready to run the world’s most advanced AI and machine learning workloads.
- Key Software Components:
HPE Ezmeral is a hybrid data and analytics platform designed to drive data-first modernizations, enabling enterprises to unlock the value of their data wherever it lives. HPE Ezmeral helps enterprises to unify, modernize, and analyze all their data, applications, and infrastructure across edge-to-cloud. This allows analytics and data science teams to scale up Apache Spark lakehouses, speed up advanced analytics workflows, and streamline the integration between Data Ops and ML Ops functions.
NVIDIA AI Enterprise is a software suite that enables powerful model training and inference in edge environments, using high-performance NVIDIA GPUs to accelerate inference, scoring, testing, device monitoring, and more. With NVIDIA’s announcement of NVIDIA AI Enterprise 2.0, enterprise AI can now be brought to everyone, everywhere.
Over the past couple of years, HPE Ezmeral and NVIDIA have been partnering together to validate the major building blocks of NVIDIA AI Enterprise with HPE Ezmeral to deliver the advanced tools and hardware to meet your analytics objectives. For example, NVIDIA RAPIDS is validated on HPE Ezmeral, optimizes GPU acceleration to workloads like Apache Spark to reduce data prep from hours to seconds. PyTorch and TensorFlow integrate seamlessly with RAPIDS to eliminate the need to procure, manage, certify, and deploy different environments. And NVIDIA Triton Inference Server, also validated with the HPE Ezmeral, simplifies the deployment of AI models at scale in production, integrating with Kubernetes for orchestration and auto-scaling.
- World Class Infrastructure:
At the foundation of this stack is a broad portfolio of HPE systems that are NVIDIA-Certified. They offer workload optimized configurations that are validated for performance, manageability, security, and scalability, and are backed by enterprise-grade support to power NVIDIA AI Enterprise. These include HPE servers available with NVIDIA A100, A40, and A30 Tensor Core GPUs. The newly announced NVIDIA H100 Tensor Core GPU based on the new NVIDIA Hopper architecture with even higher compute performance will also be integrated in the near future.
- Cloud Services:
HPE GreenLake cloud services completes the picture by giving you the option to consume this entire solution as an elastic as-a-service platform that can run on-premises, at the edge, or in a colocation facility. This makes it easier and faster to get started with AI/ML projects, seamlessly scale them to production, and have a consistent cloud experience with our cloud that comes to you and your data.
HPE GreenLake cloud services for analytics in action
Below is what HPE GreenLake cloud services looks like in the real world.
Consider a healthcare network with hospitals spanning multiple geographies and thousands of image capturing machines like MRIs and x-ray systems. Their goal is to catch abnormalities early. To do that, each hospital and clinic location has several rooms with specialized equipment and highly skilled doctors and nurses. These x-ray and MRI systems are enhanced with AI so staff can apply pattern recognizing filters to patients’ images in real time. All these locations and devices have to talk to each other and rapidly process massive amounts of information to deliver insight to doctors, nurses, and patients.
In an HPE GreenLake edge-in distributed analytics stack, HPE Ezmeral works at the edge, the data center, and the public cloud to connect the data, infrastructure, and applications from edge to cloud. By virtualizing these stacks with Kubernetes and levering our HPE Ezmeral Data Fabric, HPE Ezmeral allows data professionals to manage their data, analytics, and device monitoring from a single pane of glass.
In this modernized environment, system, patient, and diagnosis data is kept in a distributed data lake so all data is available across the entire hospital. Each in-hospital data lake is then coordinated with the centralized data lake. All data needed across edge locations are kept in this central location to ensure every hospital in the chain can benefit from updated information learned from the entire fleet. The larger compute cluster in the data center is the core integration point for developing and training new models, conducting fleet management of the over 100,000 devices in the field, and constantly monitoring the models to improve accuracy and shorten detection times. New versions of the neural nets needed for in-hospital inference engines are then shared with edge locations. This constant improvement across the entire fleet of hospitals and clinics continues to increase diagnosis accuracy and ultimately, helps save lives.
This is the HPE edge-in strategy in action, with the flexibility to pivot to follow your ever-evolving data and analytics requirements.
Better together
Coupled with NVIDIA software for DevOps engineers and data scientists, HPE Ezmeral makes the end-to-end go faster, simplifies the handoff between data personas, and makes the data more cohesive--bringing together locations into one logical set of data and applications.
With HPE Ezmeral and NVIDIA, now data scientists and engineers can confidently build for faster analytics at scale, at lower cost, and at the speed of AI.
Learn more about our collaboration with NVIDIA and the exciting possibilities of HPE Ezmeral and NVIDIA AI Enterprise: https://www.youtube.com/watch?v=Q8gQnZ69St4
Also, register for free to attend an HPE-led session at NVIDIA GTC on Optimize Performance and Management of Apache Spark Containerized Workloads.
Matt Hausmann
Hewlett Packard Enterprise
twitter.com/HPE_Ezmeral
linkedin.com/showcase/hpe-ezmeral
hpe.com/software
Matt_Hausmann
Over the past decades, Matt has had the privilege to collaborate with hundreds of companies and experts on ways to constantly improve how to turn data into insights. This continues to drive him as the ever-evolving analytics landscape enables organizations to continually make smarter, faster decisions.
