Skip to content
Knowledge Hub/Article
Article

What Is Edge Computing? Architecture and Use Cases

We explain edge computing, how it differs from fog and cloud, its architecture, 5G and MEC, edge AI, and how to secure and run edge sites.

Oğuzhan Gerçek··10 min read
What Is Edge Computing? Architecture and Use Cases

Short answer: Edge computing processes data close to where it is produced or used, instead of moving it to a distant cloud or a central data center. The goals are to cut latency and the amount of data crossing the network, to keep data on site when needed, and to keep working when the connection drops. The processing can happen on a sensor, a gateway, a local server or a single-rack micro data center. Edge does not replace the cloud: time-sensitive work stays at the edge, heavy computation stays in the cloud. The price is running many scattered locations securely.

What is edge computing?

In Red Hat's definition, edge computing is computing that takes place at or near the physical location of either the user or the source of the data. IBM describes it as a distributed computing framework that brings enterprise applications closer to data sources such as IoT devices or local edge servers.

The US National Institute of Standards and Technology (NIST), in its March 2018 publication SP 500-325, defines the edge as the network layer encompassing end devices and their users; its example is local computing capability on a sensor or a meter. Our glossary has a short definition of the term.

Why process data at the edge?

According to ASHRAE's September 2020 technical bulletin, edge computing is growing on the back of workloads whose data set is too large, or whose application is too latency-sensitive, to be sent back to a centralized cloud server. The main reasons are:

  • Latency: The planning value in ITU-T Recommendation G.114 for optical fiber is 5 microseconds per kilometer (marked as provisional), so a 1,000 km path adds 10 ms of round-trip delay from the cable alone.
  • Bandwidth: In Cloudflare's example, dozens of cameras stream raw video to the cloud for motion detection. Once each camera runs the detection on its own internal computer, most of the footage never travels to the cloud.
  • Data sovereignty: According to IBM, keeping data local makes it easier to meet compliance requirements and data sovereignty obligations. We discuss the KVKK (Türkiye's personal data protection law) and cross-border transfers in a separate article.
  • Intermittent connectivity: According to Microsoft, edge computing makes it easier to use data collected at remote sites with intermittent connectivity, such as a fishing vessel in the Bering Sea.

Edge vs. fog vs. cloud

According to NIST SP 500-325, fog computing is a layered model: fog nodes sit between smart end devices and the centralized cloud, and can be physical components such as gateways, switches, routers and servers, or virtual ones. The document's annex draws the line this way: fog is hierarchical and, in addition to computation and networking, also addresses storage and control, while edge computing tends to be limited to a small number of peripheral devices. NIST treats fog as an extension of the cloud model. The OpenFog reference architecture, for its part, was published as the IEEE 1934-2018 standard in August 2018.

The relationship with the cloud is a division of labor, not a rivalry. According to Microsoft, most organizations combine the two, with the cloud handling large computations and storage and the edge taking care of time-sensitive processing near data sources. We cover the cloud in our cloud computing article and running both together in our hybrid cloud article.

Edge architecture: from sensors to edge cabinets

IBM lists these building blocks:

  • Edge devices: Hardware that collects and processes data at the source, such as IoT devices with sensors, industrial robots and smartphones.
  • Edge gateway: A router, server or SD-WAN device that acts as a secure intermediary between edge devices and the cloud.
  • Edge servers and clusters: Local processing and storage for workloads that need low-latency responses (see our server and HCI articles).
  • Cloud or central data center: Larger workloads and deeper analytics.

ITU-T Recommendation L.1307, published in March 2024, describes micro data centers as compact, modular solutions that provide processing, storage and networking capabilities, often deployed in homes, offices, factories or base stations. Their key requirements include stable power, cooling, physical security and monitoring. A micro data center that fits in a single rack is also known as an edge cabinet, and ASHRAE's examples include phone-booth-sized enclosures that hold a single rack. According to the bulletin, the small footprint and reduced insulation of such sites disproportionately increase the thermal impact of outdoor conditions, such as heat from the sun. We cover the facility side in our data center article.

Central management: Kubernetes at the edge

According to IBM, large organizations can have thousands of edge devices, which makes deployment, provisioning and monitoring harder. The orchestration we describe in our Kubernetes article reaches the edge in two ways:

  • Lightweight distributions: K3s describes itself as a fully compliant Kubernetes distribution in a single binary of less than 100 MB, and lists edge, IoT and air-gapped environments among the places it suits.
  • Cloud-edge frameworks: KubeEdge extends Kubernetes orchestration and device management to hosts at the edge, and aims to keep edge applications running when the cloud-edge network is unstable or the edge is offline. The CNCF announced the project's graduation on October 15, 2024.

Use cases

  • Manufacturing and OT: According to AWS, manufacturers use edge computing to enable automation and collect data on site. NIST's SP 800-82r3, published in September 2023, highlights the unique performance, reliability and safety requirements of operational technology (OT), the programmable systems that interact with the physical environment.
  • Retail: In Microsoft's example, store data turns into immediate action: digital signage updates automatically, and checkout systems open new lanes when wait times grow.
  • Healthcare: According to AWS, edge devices monitor critical patient functions such as temperature and blood sugar levels, and storing patient data locally improves privacy protection.
  • CDN: According to Cloudflare, code running on a CDN edge network can customize how content is cached; we explain edge servers in our CDN article.

5G and MEC (Multi-access Edge Computing)

According to ETSI, MEC offers application developers and content providers cloud-computing capabilities at the edge of the network, characterized by ultra-low latency, high bandwidth and real-time access to radio network information. MEC is not limited to mobile networks; it also covers fixed and WLAN access.

ITU's November 2017 Report M.2410 sets the minimum requirement for user plane latency in 5G (IMT-2020) at 1 ms for ultra-reliable low-latency communications (URLLC) and 4 ms for enhanced mobile broadband (eMBB). That figure, however, covers only the radio network's contribution; if the application server is far away, the rest of the path comes on top. MEC shortens that extra path by moving the application to the edge of the network.

Edge AI

According to IBM, edge AI means deploying AI models directly on local edge devices, with no round trips to a distant server. Training usually takes place in the cloud or a centralized data center, while the trained model's inference engine runs at the edge; a retrained model then replaces the inference engine at the edge. The devices can process data with or without an internet connection.

Security and operations

According to Red Hat, the physical security of edge sites is often much lower than that of core sites, and they usually have limited or no on-site technical expertise, so the infrastructure should be easy for non-technical local staff to fix and manageable from a central location. Three measures stand out:

  1. Remote monitoring: ASHRAE calls for remote monitoring of temperature, humidity, corrosion rate, IT equipment health, physical intrusion, water intrusion and fire. At sites without on-site service, travel time lengthens mean repair times.
  2. Secure updates: Under NIST's May 2020 IR 8259A core baseline, an IoT device's software should be updatable only by authorized entities through a secure mechanism. SP 800-82r3 recommends deploying security patches in OT as quickly as possible after testing them under field conditions (see our patch management article).
  3. Zero trust: Under NIST's August 2020 SP 800-207, zero trust grants no implicit trust to assets or user accounts based solely on their physical or network location (see our network security article).

We handle the facility side of edge locations in our datacenter and infrastructure engineering service, their connectivity in network and traffic engineering and their day-to-day operation in infrastructure management.

Frequently asked questions

What is edge computing in simple terms? It means processing data close to where it is produced or used, instead of sending it all to a distant cloud or a central data center. The aim is lower latency, less network traffic and continued operation when the connection drops.

What is fog computing, and how does it differ from edge computing? Fog computing is a layered model built from fog nodes that sit between end devices and the cloud. According to NIST, fog is hierarchical, while edge computing tends to be limited to a small number of devices.

Will edge computing replace cloud computing? No. Time-sensitive processing happens at the edge, while large computations and storage stay in the cloud; most organizations use both.

What is edge AI? It means running AI models on edge devices such as sensors, IoT devices or security cameras instead of in the cloud. The model is usually trained in the cloud and runs inference at the edge.

What is an edge cabinet? It is a micro data center that brings servers, storage and network equipment together with power, cooling and physical security in a single rack. It also needs remote monitoring.

Sources