A primer for edge infrastructure deployments | TechTarget
The edge computing market is growing rapidly, driven by the need for real-time analytics, new edge technologies and rising costs in the public cloud.
But what is edge? There are many definitions of the edge, from remote offices and manufacturing sites — sometimes called near edge — to communications towers, IoT devices and autonomous vehicles — sometimes called the far edge. Uses and workloads vary, including typical office productivity, collaboration apps and VDI.
To edge or not to edge?
The key edge questions for IT are the following: Which applications should they run at the edge, which ones can they efficiently run in a public cloud and which should they run from a centralized data center? How does this affect the type of infrastructure required? The answers lie in the data:
- Data movement is expensive and time-consuming. The larger the volume, the more cost-effective it will likely be to store and analyze it where users create or ingest the data.
- Data security is critical. This has been seen in frequent attacks on data. Edge sites may be more remote and vulnerable targets, yet privacy regulations may require data to be stored locally or at least in-country, necessitating strong digital and physical monitoring.
- The true value of data lies in analysis for actionable and timely insights. Organizations can act on site-specific analytics immediately, while broader insights may come from post-processing in the cloud.
Video surveillance among top edge uses
Additionally, AI is a key emerging edge use. It makes sense to process and store data coming in from edge devices and run inference analytics locally before transferring results back to a core data center or cloud. Examples include image recognition for defect detection on a manufacturing line, virtual customer experiences in a retail location or video surveillance to ensure worker safety in a warehouse. All of these require real-time edge processing and analysis.
To illustrate, real-time video analytics in edge surveillance applications is expected to provide untampered and accurate results, such as the warehouse safety surveillance example in Figure 1. Data collection, storage and processing requirements should be specified within the application environment itself instead of applying general policies suited for cloud service providers or other on-premises infrastructures.
In a real-time AI warehouse safety surveillance operation, performance metrics include accuracy of the AI inference model, real-time processing of the model with determined outcomes and system latencies that may affect time to outcome. To optimize these metrics, compute, memory and storage resources should reside at the same location where the data is captured, such as within a video surveillance system. Figure 1 shows an example deployment, where an organization deploys a network of video cameras in a warehouse safety application to avoid forklift injury.
In a video analytics edge deployment, compute, memory and storage resources need to be dynamic and scalable, as well as meet the performance goals of the application. At the edge data collector network — in this case, video cameras — sufficient embedded compute and memory resources need to be available for each camera and at the edge processing device.
Infrastructure approaches include HCI
Edge aggregation points may have physical location constraints, such as accommodating one server rack, a closet server or even just a device. Such constraints may benefit from the tight coupling or consolidations of compute and storage resources in one server or appliance.
While there is a tendency to aim for low-cost infrastructure at the edge, especially when many edge locations are involved, low cost can mean low capabilities. It is important for IT to scale infrastructure to the workload requirements in terms of compute and storage. The good news is that newer infrastructure can deliver faster ROI in terms of efficiency in performance, as well as space and power requirements.
Management of infrastructure at the edge should support remote deployment and management. This enables workloads and infrastructure at sites that do not have local IT staff, and it ensures consistency across multiple remote sites for stronger security and reduced Opex.
One architectural approach that has become popular at the edge is hyper-converged infrastructure (HCI). It consolidates compute, network and storage onto each server for a smaller footprint that requires less space, power and cooling than more traditional IT infrastructure options.
Overall, it’s useful to keep these considerations in mind when selecting infrastructure for the edge:
- Select a cost-effective architecture that can run a broad array of workloads. Evaluate the workloads you plan to run at each edge location based on volume of data, ingest point and desirability of site-specific insights.
- Assess the processing and storage needs of those workloads. Ensure adequate compute and storage in the cluster to support current needs. Many HCI products, for example, offer edge-optimized deployment options with fewer nodes in a cluster. They use remote witness servers for reliability or enable clusters to be extended across multiple locations to share resources.
- Determine the conditions of the infrastructure location. Is it a dedicated closet or mixed-use cabinet? How well is temperature controlled? Ruggedized hardware is available from some server OEMs, where required.
- Plan for remote manageability. Many edge platforms have enhanced their remote deployment and management capabilities in recent generations, enabling a consistent provisioning process to be replicated across multiple edge locations.
The best infrastructure for your edge locations may vary. As edge uses and workloads grow and evolve — including AI, vertical-specific workloads, and standard office and remote work applications — it’s essential to evaluate infrastructure options to ensure they will meet the requirements for secure and capable compute, network and storage to support expanding operations.
About the authors
Christine McMonigal is chairwoman of the Storage Networking Industry Association’s (SNIA) Networking Storage Forum (NSF). David McIntyre is marketing chairman of SNIA NSF.
SNIA is a not-for-profit global organization made up of member companies spanning the storage market. SNIA’s mission is to lead the storage industry in developing and promoting vendor-neutral architectures, standards and educational services that facilitate the efficient management, movement and security of information.
SNIA’s NSF educates and provides insights and leadership for applying networked storage and associated technologies to a broad spectrum of end-to-end services. This expert community covers topics as diverse as block, file and object storage; virtualized and hyper-converged platforms; and cloud-to-edge infrastructure insights, accelerators and supported programming frameworks.