The View from the Edge

LM: "Edge computing" is a term with many definitions. What does Edge mean to you and why is your definition more relevant than others?

Matt Trifiro: For the first State of the Edge in 2018, we identified a need to present a framework in which the industry could have productive discussions around Edge. One of the big challenges at the time (and, to some extent even today), is the proliferation of definitions for "Edge." If you look closely, most Edge definitions are driven by a vendor or pundit's self interest, using language as an attempt to differentiate their products or points-of-view rather than as a way to bring the industry together. For example, what is the "telco Edge"? Or, what is the "IoT Edge"?

We approached the problem like lexicographers and surveyed the market. We realized, of course, that there are lots of Edges, but we also realized we had an opportunity to create clarity and offer a framework that could be independent of any particular vendor's business. Our framework has three principles:

• The Edge is a location, not a thing.
• There are lots of edges, but the Edge we care about is the edge of the last mile network.
• This Edge has two sides: an infrastructure edge and a device edge.

In 2018, we offered these principles as a gift to the industry and the industry responded by welcoming them into the vernacular. In fact, this description of Edge has been canonicalized in the Open Glossary of Edge Computing, which is an open source project at The Linux Foundation. Anybody can make comments, suggestions, and improvements, which gives us a corpus of knowledge that can evolve for the whole industry.

LM: According to estimates, the industry is expected to spend US$700 billion dollars in the next 10 years in the Edge sector. Who is spending the money and for what?

MT: This is precisely why we have been so meticulous in creating a standardized definition of Edge, so that when we do something as important as discuss a market projection, we do it using a carefully scoped model. The number US$700 billion is what we came up with in our model, given the relatively narrow constraints we put on our model, which is focused only on infrastructure spending – and does include things like economic impact. I've seen Edge projections that trumpet over a trillion US dollars, but they are including everything in that model, such as the jobs and businesses that emerge from Edge technology.

I'm not saying these other models are wrong. My point is only that we need to be clear about what we are and are not including in the projection.

Our US$700 billion projection estimates the total dollars that will be spent between today and 2028 on a very specific component of Edge computing: the infrastructure required to support Edge use cases. Not devices. Not economic impact. The report forecasts Edge data center and IT infrastructure investments in terms of capital expenditures (measured in US dollars) and the power footprint of the IT infrastructure deployed (measured in Megawatts). The predicted capital expenditures offer insight into the market opportunities for Edge IT equipment and infrastructure providers. It does not, for example, predict the opportunities for device manufacturers or auto makers, or even the economic impact of the Edge. Those are for future reports.

LM: An example in the white paper "State of the Edge 2020" cites the control of drones as an Edge use case. Isn't that also an example of how far away Edge is at the moment? Even the fastest mobile networks would not be able to control drones today – for reasons of latency, security, and safety. So how far away, in years, is an Edge scenario?

MT: Autonomous drones are a lot closer than you think. Vapor IO, my company, has a lab in our Chicago deployment that can demonstrate autonomous drones running over a wireless network. The wireless 5G technology exists today; it's just in the early stages of rollout. With drones, you can align the wireless upgrade along specific transport corridors, which is already happening in some areas. Given that the FAA continues to move quickly to support drones without human pilots, we'll see these capabilities being built out this year in certain cities to great economic benefit. Autonomous drones will create billions of dollars of economic value.

Also, Edge is not just a wireless technology. Many of the existing wireline networks, including cable and fiber, already have the latency and security problems solved for the access network (the link between the infrastructure and the devices), so it's just a matter of deploying the local IT infrastructure to support it, at the infrastructure edge, which is starting to happen at scale in 2020.

LM: We view nuclear power plants with skepticism, and rightly so: Despite high safety standards, several accidents have occurred in the past. Hospitals have to switch to paper and pencil if a Trojan encrypts their infrastructure. If Edge now connects all the daily devices and objects we use and the Internet (partially) fails or is attacked: Won't we end up back in the stone age, with more severe consequences than we face now?

MT: There are risks inherent in any new technology or infrastructure, and we have to be cautiously optimistic about how we tackle these problems and ensure we minimize or eliminate any single point of failure, but it's clear to me that Edge infrastructure will meaningfully increase the resilience of the Internet and connected devices specifically because it is decentralized, which means you can distribute risk across many different sites and have no single point of failure. For example, with seven Edge micro data center facilities in a metropolitan region, you can deliver something like twelve 9s of reliability using highly-available software failover. Barring some catastrophic event that takes out an entire city, this will greatly increase the resilience of our networks and our applications that depend on them.

The use of infrastructure Edge computing, specifically, can do much to enhance the security and resilience of our entire connected economy. Each infrastructure edge site can host software and hardware solutions dedicated to providing next generation security services at the very edge of the network. Threats that are detected in an edge location can be quarantined in that location, preventing them from penetrating deeper into the infrastructure or to the Internet.

LM: In the end, doesn't Edge, despite its decentralization tendencies, lead to an overall further centralization of the Internet?

MT: I don't agree with this conclusion. Edge, by definition, is decentralized and it is massively so, especially when compared to the Internet we have now. Today, for example, in the United States, if you want to provision a workload on Amazon Web Services, you can spawn an EC2 instance in basically two locations: US West and US East. That is extremely centralized. Now imagine a future Edge, where there are thousands (not dozens) of AWS data centers all over the world. In this future, you should be able to provision a workload not just in US West and US East, but also in Chicago West and Shanghai East.

The data paths of the Internet will also continue to decentralize. Today, Internet traffic typically passes through a small handful of large interconnection points that have emerged in hub cities, such as Atlanta and Dallas. Today's Internet passes through a relatively small number of these hubs. For example, in the US, there are maybe a dozen meaningful interconnection and exchange points. As Edge infrastructure gets deployed, a network of smaller edge exchanges will emerge to exchange data between local networks without needing to take a longer route that passes through a hub city. This creates a decentralized Internet beyond what we have seen.

LM: In an Edge computing scenario, the necessary calculations take place directly on the end devices, which have only limited capacity. At the same time, technologies like AI are very resource-intensive. Aren't these two trends opposing and contradictory?

MT: No. In a world where our devices are ubiquitously connected to low-latency networks, we can make decisions and tradeoffs about where different services and workloads will run, and this may vary with the application or even the time of day. For example, you can build an expensive sensor with a lot of AI capabilities on board. Or, you could offload that work to a nearby Edge server and deploy lower cost devices. There will be a continuum that stretches from the centralized core all the way out to the device, and developers will have the choice of where to run their workloads along the entire spectrum.

LM: Edge should work as smoothly as the power grid. But even that fails from time to time – and quite frequently in some areas. How can a much more complex Edge technology, which already requires constant software updates, guarantee a safe and secure infrastructure in the future?

MT: The secret to resilience in an Edge world is distributed infrastructure and highly available software. A properly deployed Kubernetes service, for example, can be placed in an edge data center, and if that edge data center goes down or there is a fiber cut, the orchestration engine can restart that service in another nearby location. Also, modern software techniques, such as continuous integration, automated testing, blue/green deploys, and rollbacks can all help ensure a highly-reliable system. Moreover, because most edge data centers are remotely operated, they often have sophisticated arrays of sensors that can be remotely analyzed and monitored to perform fine-grained optimizations around energy usage.

LM: A problem of today's 3G and 4G networks may remain even with Edge: There is little incentive to extend the associated infrastructure to rural areas. Moreover, parts of the necessary infrastructure are hardly profitable. What incentives can there be for looking after middle-mile architecture and edge data centers in rural areas?

MT: This is a legitimate concern, and I suspect we may want to consider solutions that have worked in the past. In the US, we have this concept of universal service. In 1935, the US established the highly successful Rural Electrification Administration and then, in 1944, extended that model to the Rural Telephone Administration. These entities stimulated rural infrastructure upgrades with programs such as offering long-term, low interest loans.

Perhaps future programs such as the Connect America Fund (CAF) will seek to include Edge.

LM: Another problem is the different speeds and budgets in nation states, countries, and regions, even cities. Certain mobile Edge applications require a consistently functioning infrastructure, such as cars, ships, and airplanes. Will the autonomous vehicle stop at the national border? Will there be a uniform Edge infrastructure, or will everyone do their own thing, as the electric car manufacturers are doing today?

MT: My crystal ball doesn't go to that level of granularity, but I will say we've had these sorts of problems in the past, and we've mostly managed to solve them. It's absolutely astounding to me that I can carry a cell phone around the world and have it just work on any foreign network – or that I can dial a seven-digit number with a three-digit country code and reach someone on the other side of the planet. We've seen successful models of compatibility and integration emerge in the past, and we should try to model those.

LM: If Edge comes in the form it's designed to take: Are there actual projects and practical considerations on how to make Edge's enormous energy footprint environmentally friendly?

MT: Environmental friendliness is one of the largest trends in data center design today. Modern data centers can often operate at ambient temperatures ("free cooling") in many geographies and for an ever-increasing number of months throughout the year. New cooling technologies use self-contained water and coolant systems, so you don't need to ingest outside water. And so on.

Ultimately, as a world, we are going to need to take alternative power generation more seriously. Fossil fuels are a problem for every industry, not just Internet infrastructure.

LM: We often talk about security, but rarely about safety. Everyday technologies on which human lives depend often have to pass complicated tests to be considered safe. What role does safety play in the development of Edge technologies?

MT: Edge applications that place human life at risk will have to go through some of the same rigorous test phases and also be subject to regulation, just like non-edge applications. I don't think there is anything particularly unique about Edge when considering this challenge, but when viewed as part of the core-to-Edge continuum, a collection of Edge data centers can offer real-time collaborative monitoring and processing for Edge devices, and this stands to improve the overall safety of applications and devices that impact our physical world, such as vehicles.