AUSTIN, Texas – Much of the discussion of edge computing focuses on latency, and the need to move content closer to users to improve the experience for streaming video and gaming. Amid an explosion of consumer mobile devices, edge infrastructure holds the promise of dramatically improved performance, reducing “lag” for online content and applications.
But the disruptive nature of edge computing could go well beyond latency. Several startups view edge data centers as part of a fundamental shift in Internet architecture, moving network interconnection points – the key intersections that allow data to move between networks – from the core of the Internet to its outer perimeter.
Edge computing startups like Vapor IO and EdgeMicro envision a future in which thousands of small data centers provide multi-tenant colocation services at telecom towers and urban rooftops. That includes peering and interconnection, which has historically been focused at the core of the network, in the largest buildings in major business markets.
As he stood on the stage at the Edge Congress 2018, with the global leadership of the nascent edge sector seated before him, EdgeMicro Chief Technology Officer Josh Snowhorn laid out the challenge. Austin is a technology hub, home to Dell and a vibrant startup ecosystem.
“This city becomes the edge, because we don’t have an interconnection facility here,” said Snowhorn. “If you want to interconnect, you’re going to Dallas.”
Thus, sending data to someone across the room in Austin could require a 360-mile round trip to Dallas. It’s an inefficiency that gets even more glaring as computing and data storage become distributed. That’s why the rise of edge computing must involve new approaches to Internet architecture.
“There has to be peering,” said Cole Crawford, the CEO of Vapor IO, which is based in Austin. “Peering is where all this starts. Automation is where it ends. Programmability is the end game.”
The Edge Flips the Model
That vision would transform the performance of the Internet, creating local connections that could dramatically reduce latency. Crawford’s vision for a software-defined edge fits with plans for 5G wireless connectivity, as mobile network operators tap network functions virtualization (NFV) to create a flexible, configurable network.
That’s a lot of change. Many of the conversations at Edge Congress, a new conference series from Broad Group, sought to address the challenges in making this vision a reality. It will require lots of infrastructure, lots of software, and new business models that will disrupt the status quo.
“There is a tremendous difference between operating a large site, and taking that same IT load and distributing it across 1,000 sites,” said Ty Schmitt, a vice president and fellow at Dell EMC. “The data centers of today define the applications that go into them. The edge flips that. The application defines the hardware. It’s gonna be a chaotic, fun ride.”
“The data centers of today define the applications that go into them. The edge flips that. The application defines the hardware.”
Ty Schmitt, Dell EMC
“This is just the beginning,” said Hugh Carspecken, CEO of DartPoints, another provider targeting the edge data center sector. “It’s a blank canvas. There’s a lot of talk. In a couple of years, you’ll see fairly large facilities in places we never thought.”
Moving Intelligence to the Edge of the Network
As we’ve previously reported, edge computing is not a single technology, but a phrase that describes several layers of infrastructure, some of which are refinements of existing models. Edge infrastructure is a response to new technologies – such as autonomous vehicles and distributed AI applications – that require low latency and close proximity to users. These technologies are shaping the future of Internet infrastructure,
The evolving edge will include telecom infrastructure for the rollout of the new low-latency 5G wireless network, as well as data center infrastructure to house servers and storage hardware for applications – as well as the analytics to enable a new layer of intelligent, real-time services.
“This is where edge creates value,” said Jack Pouchet, VP of Business Development at Vertiv. “We need intelligence as close as possible to where this data is being generated. Just because a quadrillion gigabytes of data are being created, that doesn’t mean we have to store it all. We need to get data to the right place, at the right time, and act appropriately on it.”
From Carrier Hotels to the Edge
Peering at the edge would mean rethinking a core principle of interconnection, which has historically focused on the densest concentrations of networks. To understand the shift, a little history is in order.
The Internet is a network of networks, and requires physical connections between these networks for data to reach your computer. That’s why network connections in their many forms – in meet-me rooms, internet exchange points, peering centers or across software-defined network (SDN) connections – play an important role in making the Internet work.
In the early days of the web, this happened in carrier hotels, large multi-tenant buildings housing dozens of telecom providers and ISPs. These buildings usually were often skyscrapers in the central business districts of major cities (think 60 Hudson or One Wilshire) that operated a “meet-me room” for providers to make physical cross connects with other networks.
“We need intelligence as close as possible to where this data is being generated. This is where the edge creates value.” Jack Pouchet, Vertiv
The debut of Equinix in 1997 shifted the geography of these key network intersection points to suburbs such as Ashburn in Northern Virginia and Palo Alto in Silicon Valley. These locations benefited from a network effect, as each new connection made the sites more attractive to networks and content providers.
In recent years, regional companies like Cologix and DataBank have worked to create interconnection ecosystems in smaller cities, while Open-IX has worked to support extend interconnection to more data centers through standards and certification.
Rethinking the Mobile Model
Edge peering at tower sites would also mark a shift in how these locations have managed data traffic. The major tenants at towers have been mobile network operators (MNOs), whose antennas provide wireless service for cell phone users. From the tower, each MNO moves data from its base station back to its wireline network, a process know as backhaul. In the mobile world, data flows from the traffic to the core before it finds its way to its destination.
“The mobile guys today don’t really peer,” said Snowhorn. “They backhaul all their traffic.”
“The data center ecosystem has developed around this neutral platform,” said Sharif Fotouh, of Compass EdgePoint, which is building a suite of hardware and software to support edge deployments. “The network and facilities need one another. I don’t see that same balance on the edge.”
“Ultimately, the network is not out at the cell tower,”said Raul Martynek, the CEO of DataBank. “It’s a backhaul point. There’s interim points between Ashburn and the cell tower, and history tells us those interim points will happen first.”
But for startups like Vapor IO and EdgeMicro, the fullest expression of the edge is a network that can exchange packets at the edge. “We’re just moving the carrier hotels closer to the eyeballs,” said Crawford.
“Our goal is to put thousands of exchanges out there,” said Snowhorn. “Our goal is microsecond latency within an individual market.”
Both companies have developed software to manage these connections at edge locations:
- Vapor IO ‘s Kinetic Edge uses software and high-speed connectivity to bring multiple micro data center facilities into a single virtual facility with multiple availability zones. Modules are meshed together with high-speed fiber across a group of data centers. The design builds upon the “availability zone” approach popularized by Amazon Web Services, creating a cluster of local data centers to easily hand off data and traffic.
- EdgeMicro’s Edge Traffic Exchange moves local IP access between the cell tower radios, through common telecommunications switching and routing equipment, and into the server and storage equipment within its modular data centers, without transiting the signal back to a fixed data center.
Companies in the interconnection market, including Equinix and EdgeConneX, are watching the opportunity closely. Some veterans in the interconnection space believe distributed exchanges are inevitable.
“5G will change a lot of business models,” said Ivo Ivanov, the CEO of DE-CIX, at the company’s recent media day. “We will probably need Internet exchanges at every crossroad and every tower. The interconnection solution should get closer to the user – and their car. We’ll see how the market accepts this and validates this.”
Reducing Cost Through Shared Infrastructure
So what about the economics of this model? There are two factors to consider: the economics of thousands of data modules at edge locations, and the economics of interconnection at these sites.
If there’s one thing most edge technologists agree on, it is that success will require aggressive steps to reduce the cost of deploying infrastructure. This is driving the focus on standardization, including repeatable designs for modules and the use of “white box” servers, which is a major focus for AT&T in its push to 5G and network virtualization.
The edge data center startups believe this focus on cost will disrupt the practice of single-tenant, dedicated infrastructure that has dominated the mobile networks. Interconnections between networks at the tower, rather than the core “creates an efficiency, where you can interconnect at the edge, and not backhaul to an MSO or cloud,” said Snowhorn.
“A lot of the models at the edge are around shared infrastructure,” said Phil Kelley, SVP of Corporate Development and Strategy at Crown Castle, a major developer of wireless towers.. “That’s how you drive the costs down.”
“Shared infrastructure is going to happen at the edge,” said Crawford.
What’s the business model for interconnection at the edge? The economics of interconnection are not always simple. Internet exchanges can operated by building owners, commercial service providers, and non-profits. Some cross connects and peering relationships are free, while others involve payments.
In recent years, the emergence of software-defined networking has provided network capacity that can be configured through a web portal, dramatically simplifying the way data center customers manage their connectivity. SDN figures to be a key component of edge for both the colo operators and wireless networks.
Will mobile network operators embrace multi-tenant infrastructure at the edge? What kind of revenue will edge colocation and interconnection create? How fast will these ecosystems develop and grow? Do current business models in core Internet markets offer insights into economics at the edge? The next three to five years will provide answers to these questions.
“These are really exciting times,” said Pouchet. “There’s opportunity for new business models.”
Explore the evolving world of edge computing further through Data Center Frontier’s special report series and ongoing coverage.