The Rise of Edge Computing

Over the last several years, there was a sea change in technologies and the way we use them. In a world where people and things are always connected, and with 75% of data expected to be created at the edge within the next few years, it makes sense to take the compute to where the data is, i.e., to the edge. Gartner defines edge computing as solutions that facilitate data processing at or near the source of data generation. Any application that needs to respond quickly to user requests or events is a candidate for edge deployment. These include autonomous vehicles, industrial process control, and even digital distribution of streaming content. In this narrative, we will explore why edge computing is on the rise and its relationship to the cloud, what the industry is doing about it, and how to keep ahead of the rest as the edge unfolds.
WHY THE EDGE?
Déjà vu ! We have been there before. The computing paradigm evolved from centralized mainframes to distributed computing with the advent of personal computers and servers. Then came the cloud, a centralized shared infrastructure for scaling as the business grows. And now we are back to distributed computing, this time at the “edge.”
Artificial intelligence (AI) apps, like facial recognition check-ins at airports, self-driving cars, expert advice apps in healthcare, or digital distribution of streaming video, need large amount of sample or historical data to build a decision model, which is ideally suited for cloud data center processing with lot of compute power and storage. However, when it comes to putting the models to use, the locally generated real-time data would have to be sent to an edge node for processing and get a quick response indicating a decision or a course of action. As Peter Levine, General Partner from Andreessen Horowitz, puts it, the cloud is about “learning” and the edge is about “action.”
For the action to be timely or meaningful, one needs to be able to convey large amount of data quickly to an edge compute node. This requires high bandwidth and low latency networking, and very often, a mobile network.
Enter 5G! The 5G wireless interface provides high capacity. In addition, if an edge computing node is at a cell tower, mobile switching center, or at ingress to carrier’s core network, one can achieve extremely low latencies. This can be thought of as Network Edge Compute (NEC) node. In enterprise applications, where a factory floor data is collected with Internet of Things (IoT) sensors, over a local 5G cellular, WiFi or fiber, the on-premises processing node is termed as a Multi-access Edge Compute (MEC) node (although ETSI uses this term to represent network node as well). These types of architectures provide high bandwidth and low latency, well-suited for real-time applications which need to respond quickly to user requests or events.
Edge computing may be more generally called the “edge cloud,” because, in addition to compute, the edge would have storage and networking capabilities. In addition, an edge cloud also uses virtualization, workload segmentation by user or application, automation and orchestration of management and maintenance activities, and ability to shift workloads from one edge to another in mobile-centric applications.
Use Case for Digital Distribution of Streaming Content
Let’s examine the use case for digital streaming. The Content Delivery Networks (CDN) were, in a way, precursor to the edge cloud. CDNs play a critical role in improving user experience by striving to provide quick response to user requests for streaming content. Major streaming services accomplish this by either establishing their own Points-of-Presence closer to their user communities, or using a third party CDN infrastructure, or entering into commercial agreements with Internet Exchange Points (IXPs) or Internet Service Providers (ISPs) to install edge nodes to cache content closer to users. The figure below shows such an approach. Netflix, for example, delivers 90% of its traffic to its members via direct connection between its edge nodes and IXPs/ISPs.
These edge nodes store a set of files which are expected to be requested by the community they serve and report status to the cloud. When a playback application in the cloud receives a customer request, it determines the nearest edge node to the customer that has the requested content, and directs the user device (Smart TV, Laptop, Tablet, etc.) to fetch the content from that edge node.
User profiles and usage history are generally not stored in the edge nodes. In the cloud, the content consumption history of users associated with an edge node can be applied to an AI model to predict potential future content consumption. The edge nodes are periodically updated with files related to expected future content consumption, so that they are there ready for streaming on request. Multiple edge nodes may be updated with same content if it is popular or if they have an intersection of same set of users they serve.

As can be seen above, the edge node plays a rather passive role in current content streaming infrastructure. So, it is well suited for caching of static content but not for decision making. It is not optimized for mobility, as the same edge node keeps serving the content even though the mobile device that requested the content may have moved. Also, the edge itself does not utilize artificial intelligence, although selection of what content to cache where may use machine learning algorithms back in the cloud.
That is where the “new edge” comes in with powerful compute, storage, and networking capabilities that allow it to process events at the edge, utilize artificial intelligence to enhance user experience, and even to cooperate with other edges to “follow the user” for optimal responsiveness.
INDUSTRY ON THE EDGE
Before looking into what the industry at large is doing to tackle edge computing, it is worth taking note of other developments, particularly virtualization and automation.
Server virtualization and multi-tenant support for running independent workloads has been around for some time. More recently, wireless carriers also started deploying virtualized network infrastructure, replacing many dedicated hardware-based elements (e.g., routers, switches, firewalls) with virtual network functions (VNFs) running on general purpose edge computing processors. As part of managed service offerings, carriers are pushing VNFs all the way into enterprise customer locations, deploying edge capabilities close to end users.
Edge nodes need to be flexible to support different types of hardware which may be optimized for storage, compute, or networking, depending on the application. Also, they may use virtual machines, containers, micro services or serverless infrastructure as needed. Some applications may need hardware accelerators, such as Graphics Processing Units (GPUs) or Neural Processing Units (NPUs) for encryption/decryption and machine learning. As edge nodes are much more widely dispersed than cloud data centers, it is important to minimize human intervention for deployment of the software platform and applications, once the hardware is in place. So, automation and orchestration become even more critical for successful edge deployments.
Automation focuses on making a time-intensive, manual task or a group of tasks more efficient and reliable, and performed without human intervention. An example is deployment of an application instance on a virtual machine or automatic scaling by adding new instances of an application with increased traffic. Orchestration looks at the big picture and workflow across the automated tasks, and enhances security with identity and access management policies. An example is ensuring that a set of applications are executed in a particular sequence in order to provide a service.
Now, we will briefly explore industry activities related to cloud computing infrastructure and deployments. This is by no means an exhaustive survey of what is happening in the industry but just a sampling of the initiatives and offerings.
Edge Infrastructure
All the major Cloud Service Providers (CSPs), systems vendors, and standards organizations are in on the action. The CSPs, in general, are extending their cloud capabilities to customer or carrier premises in order to provide cloud-like capabilities at the edge. The systems vendors are bringing increasingly sophisticated hardware and software capabilities to the edge. The standards organizations are trying to create some structure around these efforts in order to make it easy to innovate, deploy and manage edge applications.
Amazon
Amazon has two types of edge computing offerings. They recently announced “Amazon Outposts,” that allows building an edge-cloud on-premises in customer data centers. The customers are shipped physical racks of Amazon Web Services (AWS) servers and once installed, the racks will connect back to the AWS mothership over the Internet. Through AWS Management Console, the customer can then use the compute and storage services on pay-per-use basis, just as with the cloud. Earlier, Amazon introduced Snowball Edge, similar to Outposts in concept, except that it doesn’t connect back to mothership and is designed to run VM servers and serverless applications, including AWS Lambda invocations, which are event-driven function code executions. The Amazon Greengrass addresses edge computing needs related to the Internet of Things (IoT) marketplace, by enabling a gateway and Lambda capabilities for offline local computing, including machine learning inferencing.

Amazon Nitro platform offers dedicated hardware for networking, storage, and security, along with a lightweight hypervisor. Its flexibility enables virtual machine instances to be created which can be optimized for compute, memory, storage, machine learning, etc., based on the type of application. Snowball Edge can be configured using the Nitro platform.
Microsoft
Microsoft offers a portfolio of edge products, from OS to database to AI, with initial focus on the IoT market. They are very active with more than 300 patents in the space of edge computing. Microsoft’s Azure Sphere aims to power the next-generation microcontrollers and has three components — i) chip with processor cores, wireless connectivity, and security, ii) OS based on open source Linux, and iii) secure cloud connectivity to Azure IoT Hub, which is the device management layer in the cloud.
Azure IoT Edge is a service that builds on top of IoT Hub, meant for customers who want to analyze data on devices, or “at the edge,” instead of in the cloud. Microsoft Azure also offers SQL Database Edge, which is a resource-light, edge-optimized data engine with built-in AI. It provides support for processing and storing graph, JSON, and time series data in the database, coupled with the ability to apply analytics and in-database machine learning capabilities on non-relational datatypes.

Building and deploying applications for these edge environments will involve developer tools, like containers and Kubernetes. Microsoft announced general availability of Azure Kubernetes Service (AKS), which provides a host of automated management capabilities for Kubernetes, the popular open source software that helps unify cloud and edge use cases by abstracting out the underlying hardware that hosts applications.
Microsoft announced, in March 2019, general availability of its Azure Data Box Edge and the Azure Data Box Gateway for on-premise deployment, similar in concept to Amazon Outposts. Data Box Edge, which is described as an on-premises anchor point for Azure, is offered on a pay-as-you-go basis, just like any other Azure service and the hardware is included. Azure Databox Gateway is a virtual storage device, which resides on customer premises, and can be used for transferring data to the cloud for archival, disaster recovery, or cloud scale machine learning.
Google focuses on “intelligence everywhere,” serverless scalability, and security. Machine learning has been the core of many products and services at Google. Their primary mantra is pervasive AI with end-to-end AI architecture from edge to the cloud. To accelerate machine learning apps, Google built a custom ASIC (Application Specific Integrated Circuit) called the Tensor Processing Unit (TPU). TPU is highly optimized for massively parallel calculations required by neural networks.
Recently, Google has announced the availability of Edge TPU, a miniature version of Cloud TPU designed for single board computers and system on chip devices. Google’s TensorFlow can be used in the cloud for training machine learning models. TensorFlow Lite is a flavor of TensorFlow meant for mobile devices and low-powered environments. Edge TPUs modules have a low footprint that makes it ideal for embedding them in devices such as drones, cameras, and scanners. Edge TPU complements the Cloud TPU by performing inferencing at the edge, based on trained cloud models.

Google’s IoT Edge is a software stack that extends Google Cloud’s powerful AI capability to gateways and connected devices. The IoT Core securely connects edge devices to the cloud, enabling software and firmware updates and managing the exchange of data with the cloud. By running on-device machine learning models, Cloud IoT Edge with Edge TPU provides significantly faster predictions and classifications for critical real-time applications while ensuring data privacy and confidentiality.
IBM
“IBM Edge Computing” is an advanced platform supporting real-time AI, 5G and IoT applications. It facilitates secure deployment and remote management of these applications on edge devices, servers and gateways across hybrid cloud environments. With the acquisition of Red Hat, IBM offers OpenShift Container Platform, an on-premises “platform as a service” built around Docker containers, orchestrated and managed by Kubernetes on a foundation of Red Hat Enterprise Linux. This enables IBM Edge Computing to provide a full range of deployment flexibility to manage, move and secure workloads. It helps manage edge workloads at massive scale with visibility, control and automation.
IBM Watson IoT Platform Edge brings intelligence and cognitive solutions to the edge. It enables automated deployment of services at the edge, communication of relevant messages between the edges and the cloud, and offline operation with updates and synchronization when connectivity becomes available.
IBM has also been a pioneer in “blockchain” technology and its application in various vertical markets. For example, IBM Food Trust, a blockchain-based network of players in the food supply chain, from producers to manufacturers to retailers, enhances food safety, and transparency of its journey from farm to the dinner table. IBM enables use of use IoT sensors to track fruits, vegetables, or any other food items on the long journey from field to grocery store. AI-models, along with blockchain, could also help retailers and producers learn more about consumer eating patterns, facilitating demand-supply coordination, moving the society closer to zero-waste food consumption.

IBM also supports multi-cloud environments with its Cloud Pak for Multicloud Management, which unifies cloud platforms from multiple vendors (Amazon, Microsoft, Google) into a consistent dashboard, from on-premises to the edge. In addition, IBM also brings AI tools for accelerated deep learning to the edge for visual and speech recognition, video and acoustic analytics, and creation of sophisticated enterprise applications.
Industry Stalwarts
Equipment and systems players, such as, Cisco, Cienna, Hewlett Packard Enterprises, and NVIDIA all have edge computing in their product portfolio.
- Cisco is building on its strengths in switching, routing, and networking to offer a portfolio of edge computing products. For example, Cisco Aeronet Access Point supports distributed wireless sensors and enables design of custom apps that process data from edge devices locally and send results to cloud services for further analysis. Cisco uses the term “fog computing” to denote the layer of processing between the edge devices and the cloud. They offer a Cisco Kinetic Edge & Fog Processing platform which is designed to extract data from the connected devices, apply rules and logic on data in motion or data at rest, compute anywhere in a distributed network, and move data to the various applications where it can be used to drive desired outcomes.
- Cienna suggests that the economic advantages of edge cloud arise from the shared use of generic white box resources across a range of applications, e.g., mobile apps, vCDN apps for video content and webpage delivery, IoT apps, DNS, DDOS, and vRAN apps for mobile edge. Cienna’s Blue Planet Intelligent Automation portfolio enables orchestration and management of virtualized network functions, software defined networks, various access technologies, and incorporates AI to achieve business agility and accelerate service velocity.
- Hewlett Packard Enterprises (HPE) focuses on convergence of Information Technology (IT) which deals with systems and applications, and Operational Technology (OT) which deals with control systems, industrial networks, and data acquisition. HPE Edgeline Converged Edge Systems converge IT and OT in a single, rugged system suited for harsh environments to enable innovative new abilities at the edge. The Edgeline portfolio also supports 5G and MEC applications.
- NVIDIA introduced, in May 2019, its EGX edge computing platform for performing low-latency AI on continuous streaming data from 5G base stations or other edge locations. EGX begins with the tiny NVIDIA Jetson Nano, which delivers one-half trillion operations per second (TOPS) of processing in only a few watts, and scales all the way to a full rack of NVIDIA T4 servers, delivering more than 10,000 TOPS. NVIDIA is also promoting “On-Prem AI Cloud-in-a-Box”, which combines the full range of NVIDIA AI computing technologies with Red Hat OpenShift and NVIDIA Edge Stack.
Trailblazing Startups
No business transformation discussion is complete without looking at some trailblazing startups with ambition to disrupt the status quo.
Pensando
Pensando is a venture-backed startup that came out of stealth mode in October 2019. It was started by ex-Cisco execs and its investors include Hewlett Packard Enterprise, Lightspeed Venture Partners, Goldman Sachs, and Equinix. There are bold pronouncements that Pensando is going to take on Amazon Nitro. Why? While Nitro systems use dedicated ASIC hardware cards that offload networking, storage and management tasks, it is pre-configured at the time of the order as a compute-optimized or a storage-optimized system. Pensando claims to be able to do customization and host server offload with a programmable software-defined approach and achieve up to 9x improvements in performance and scale.
Pensando powers software-defined services with Distributed Services Cards (DSCs) that deliver high-performance cloud, compute, networking, storage and security functions. They support virtual appliances at every node, eliminating the need for legacy appliances including firewalls, load balancers, and encryption devices. The DSC can be easily installed in standard servers to run any of the virtual functions and it is compatible with virtualized or bare-metal servers as well as containerized workloads. It can also be used in-line at 100G wire-speed, similar to a Network Interface Card, but versatile enough to handle both networking functions, like terminating and encrypting IoT connections as well as storage functions, such as compression, decompression and checksum calculations for deduplication of data. The DSC also supports Non-Volatile Memory Express-over Fabrics (NVMe-oF) including TCP transport fabric. This has significant implications on edge node architectures, as discussed next.

Lightbits Labs
Lightbits Labs, a startup based in Israel and silicon valley, championed the NVMe/TCP standard. While direct-attached storage architectures offer high performance and are easy to deploy at a small scale, they are constrained by the ratio of compute to storage. NVMe/TCP offering from LightLabs separates the compute and Solid State Drive (SSD) storage to make it easier to scale high-performance storage independently of compute. Unlike storage-specific network protocols, such as Fiber Channel and Remote Direct Memory Access (RDMA), TCP is widely used in data center, local area, and wide area networks. What this means is that NVMe/TCP breaks the boundary limiting an SSD to a single dedicated server. SSDs can be shared across servers over a standard high-speed TCP network.
This also means that several edge nodes may share centralized SSDs over TCP. Space may be limited at edge locations and some of them may be in rugged environments, making it difficult add significant storage on site. With NVMe/TCP, storage can be scaled as needed and accessed almost at wire speed to support data analytics and machine learning applications at the edge.
Standards Activities
There are several efforts worldwide to address various aspects of edge computing and most importantly the management and orchestration of edge deployments. There are multiple edge open source and standard initiatives (e.g., ONAP, Open Stack, ONF, CNCF, ETSI MEC, OPNFV, Open Compute Project, LF Akraino, 3GPP, etc.,) that are converging to create an ecosystem that will support edge computing and services.
The Linux Foundation is unifying a number of its projects into a new umbrella organization, LF Edge, to establish an open, interoperable framework for edge computing independent of hardware, silicon, cloud, or operating system.
Some of the projects at LF Edge are Akraino Edge Stack, EdgeX Foundry, and Open Glossary of Edge Computing, formerly stand-alone projects at The Linux Foundation.
Launched in 2018, Akraino Edge Stack aims to create an open source software stack that supports high-availability cloud services optimized for edge computing systems and applications. The Akraino Edge Stack is designed to improve the state of edge cloud infrastructure for enterprise and carrier edge networks. It will offer users new levels of flexibility to scale edge cloud services quickly, to maximize the applications and functions supported at the edge, and to help ensure the reliability of systems that must be up at all times. The initial code for Akraino Edge Stack was contributed by AT&T and Intel. The reference model is illustrated below.

EDGE CLOUD DEPLOYMENTS
Edge can be almost anywhere — a customer building, a cell tower, carrier network edge, IXP/ISP hub, central office, etc. Several players are betting on growth in edge computing applications and are starting to deploy edge cloud, based on their strengths.
5G Edge Computing White Paper says, “It is likely that the low latency — high reliability applications at the metro cloud level, and the ultra-low latency/high reliability applications at the far-edge and on-premise levels, will represent only a fraction of overall applications. However, the more specialized 5G era applications promise to be high value use cases that will drive innovative business models and new transformative value creation. An implication of this is that partnership business models between ‘application and content providers’ (ACPs) and network providers and/or edge cloud providers will become very important to realizing this new 5G era services potential.” This is true even if an application may not necessarily use 5G.
We already see such partnerships emerging. Recently, AT&T and Microsoft announced strategic partnership to bring Azure services into AT&T 5G network edge. According to the press statement, for Microsoft, this enables Azure cloud services to connect to more customers and devices across the U.S. through AT&T’s nationwide wireless network. For AT&T, the partnership helps create new ways for its customers to directly access a multitude of cloud services closer to where they do business.
Some believe that the cell towers are the next frontier for edge computing. This is boosting collaboration between data center companies and tower real estate specialists. Digital Bridge, a veteran of the tower industry, raised a $4 billion “patient capital” fund with partners, partly to be used to acquire Zayo Group (expected to close in the first half of 2020), with its substantial data center colocation platform and a carrier business. Digital Bridge, through its subsidiary, currently has more than 268,000 owned and managed sites nationwide, including 55,000 wireless and broadcast towers, as well as rooftops, land parcels, and billboards. They are exploring use cases that warrant building out edge micro-data centers at cell towers.
At CenturyLink Inc., about 100 facilities that used to store telecom equipment are now outfitted with servers. They are making these servers available to corporate customers in sectors like retail and industrial robotics. They believe that there is enough activity in this space to confidently build out this infrastructure.
Akamai introduced its Edge Cloud for the delivery of data to connected devices and in-application messaging at scale. Akamai says its Edge Cloud solution line provides global scale that other data platforms for IoT and messaging lack. IoT Edge Connect, a new product within the Edge Cloud solution line, enables IoT devices and applications to send or publish information about a given topic/event to an Akamai edge server that functions as an IoT message broker. With IoT Edge Connect, Akamai says developers can enable low-latency interactions with millions of endpoints and process data in real-time.
MobiledgeX, the edge computing company founded by Deutsche Telekom AG, is supporting prototyping of developer use cases such as live trials around Augmented Reality (AR) and Mixed Reality (MR). At runtime, MobiledgeX Edge-Cloud spins up containers on-demand in edge cloud locations (also known as “cloudlets”) that optimally fulfill the needs of the desired application and user quality of experience.
China Tower and Alibaba have announced that they formally established a strategic partnership. China Tower, the world’s largest telecom tower operator, with 1.9 million tower sites, will collaborate with Alibaba on cloud computing, edge computing, big data, and 5G, to implement cloud-based smart city and smart transportation projects.
GAINING THE EDGE
With billions of mobile devices and sensors now connected to the Internet, and with growth of artificial intelligence applications, more people and enterprises need their computing power closer to them. This creates huge opportunities for some players while potentially creating challenges to existing cloud players.
Bloomberg says, “Over time, cloud will be primarily used for storage and running longer computational models, while most of the processing of data and AI inference will take place at the edge.” They estimate the size of the edge-computing market at more than $4 trillion by 2030.
Wireless carriers and the owners of cell towers, in theory, have a big advantage in the edge-computing race. They control access to high-speed communications networks and have valuable real estate.
However, cloud computing isn’t going away anytime soon. But there’s more pressure on the industry’s leading players (Amazon, Microsoft, and Google) to team up with wireless carriers, so that they are not left out of the growing edge market. We already see this happening with the announcement of Microsoft-AT&T partnership.
Cloud computing has contributed significantly to Amazon’s earnings in recent years compared to the e-commerce business. However, some analysts believe that there’s a threat looming at the edge of the network.
There is also the contrarian view. The cloud giants are pushing the cloud functionality to the edge. They have huge customer bases and developer ecosystems, and thousands of servers across several cloud data centers. That is, they have scale. It would be a monumental effort and investment for a carrier to try to build a cloud offering to rival that scale. The carriers also recognize this. They are treading the waters carefully and trying to ensure that they broker their customers’ relationships with the cloud provides. But this may be hard as most customers already have direct relationships with the cloud providers for the services they offer.
So, the verdict is still out on who will emerge from gaining an edge. It may after all be the makers of jeans, as with the early railroad industry.