The entire argument in favor of edge computing is that it expedites processing by closing the distance between processors and the data they need. It’s the movement that’s triggered the redistribution of data centers across broad geographies and miniaturized the form factors of many servers.
But is it as simple as relocating the processor? Or is the network that ties everything together – particularly with its Internet Protocol routing that bounces signals sent between adjacent buildings clear across the planet – messing things up more than we care to admit?
What Edge, Where?
A telltale clue found in a recent AT&T test of a system for real-time tracking of airborne drones points to the likelihood that “the edge,” at least from the service provider’s perspective, isn’t where we thought it was, nor perhaps what some in the telecommunications field would like it to be. As Data Center Knowledge recently reported, that test, staged at the AT&T Foundry experimental facilities in Plano, Texas, involved an on-site data center that appeared to be fulfilling the role of a future on-site micro data center (µDC) at a customer facility.
Yet as Jeff Shafer, an AT&T assistant VP, explained to us, that data center was portraying a different role: that of a National Technology Center, or NTC. It’s AT&T’s term for an owned and operated facility on its own real estate — perhaps newly built, perhaps repurposed from an existing function — parts of which exclusively host customer data and applications.
That sounds a lot more like the network core than the network edge. Indeed, in a redacted November 2018 letter to the Federal Communications Commission [PDF] explaining the core components of its Wireless Resiliency Cooperative Framework — specifically those components threatened by 2017’s Hurricane Irma — AT&T assistant VP Joseph Marx explicitly listed an NTC.
But the company’s vernacular dates back to the days of the Bell System, when the telephone network core was maintained by “field technicians.” An NTC may be part of a department AT&T abbreviates as TFS DCOR (Technical Field Services/Data Center Out-of-Region). It’s the core from the network’s perspective, but the “field” from the customer’s point of view. That’s the distinction the company relies on when introducing its vision of the edge to prospective customers — including those considering relocating their data center assets offsite or outside of colo facilities.
“There is a whole lot less interest these days in building data centers and managing them,” Shafer told us in an interview. “In fact, many of them are looking to move out of what they have. . . Why would you want to build your own network of micro data centers all over the place when you could leverage a shared infrastructure that exists, and provides the same latency?”
In AT&T’s present network, Shafer said, data traffic that enters the wireless Radio Access Network (RAN) gets backhauled to the NTCs, with several aggregation facilities along the way. A µDC adjacent to a RAN tower, as some edge computing providers have already deployed in limited areas, is actually separated from that tower by a handful of stops, one of which is the NTC. Thus, a network connection between a RAN tower and a µDC might look like a road racing course, where the NTC serves as the tricky hairpin turn.
If that’s true, then why are µDCs continuing to be built and deployed, and why are their proprietors calling their locations “the edge?” We asked Michael Clever, senior VP and Head of Cloud RAN for the Mobile Networks division of Nokia, one of the world’s three major producers of 5G RAN equipment. As Clever stated in a note to us, Shafer’s description of AT&T’s network is indeed accurate today. But the 3GPP organization’s official specifications for 5G now include a technology for implementing local breakouts — ways to detour selected traffic to on-site facilities before it’s shuffled off toward the NTC.
“In 5G, we have network slicing and different slices may need local breakouts,” explained Clever. “These local breakouts can provide internet or intranet connectivity to far-edge sites, and hence local data processing can make sense (e.g. video pre-processing, AR). This also requires some packet core functionality onsite, of course, that can and will run at the same far edge data center.”
“Far edge” is what Nokia calls data centers adjacent to RAN towers.
One of the goals of 5G Wireless technology is to cover multiple service areas no smaller than a neighborhood or perhaps the size of a city block through the deployment of much smaller transmitters that are easier to cool and maintain. Inevitably, multiple carriers’ transmitters may find themselves sharing the tops of lampposts and traffic lights. That’s difficult enough, but the sharing of nearby processors and servers could be impossible were it not for virtualization.
Mellanox is one of Nokia’s suppliers of data center equipment, including its edge data centers. Those facilities include Nokia’s AirFrame “open edge” form factor, introduced in April 2018 and designed particularly for telco edge environments.
Recently, Nokia’s Nuage Networks venture has been partnering with Mellanox to devise a system for multi-tenancy in micro data centers. At the core of this concept is a device they call “switch-on SmartNIC,” which their engineers introduced last April at the Open Infrastructure Summit in Denver. As Nokia’s Clever told us, his company has been answering requests for far-edge data center equipment that can leverage this concept to deploy virtual base band units or virtual deployment units (vDU). This would effectively carve a micro data center into “stripes” for multiple tenants, enabling them to share both data centers and radio units without their having to be stacked atop one another like children’s blocks. “This would dramatically reduce the transport and deployment costs and also give room for additional applications,” Clever said.
But contrary to what’s been considered before, that telco cloud will not be striped for use by private customers.
“In the enterprise case, the far-edge data center can or even has to be located on a private-property environment,” he stated, “and the available compute capacity can be used to run other enterprise-specific services. Intranet connectivity is not an issue in this case, and a private cloud [would be] the usual environment. This is especially required for ultra-reliable low latency (URLL) network use cases — e.g. for fully automated factories, autonomous driving, etc. This is one of the key use cases for 5G in the 3GPP standards as well.”
So, it’s looking less likely, at least from Nokia’s perspective, that “far-edge” deployments adjacent to RAN towers will be leveraged for use with private customers’ edge clouds. Is that because telco-owned facilities simply aren’t close enough to customers for latency to be lowered for everyone? Is it due to what AT&T’s Shafer pointed out — the traffic route is too circuitous or hair-pinned to be efficient? Or is there an infrastructure issue about which we haven’t been made aware?
“It is about the need to stay in control of business-critical processes,” responded Clever. “Enterprise customers often do not accept mission-critical processes and information outside their own control/premises for a number of reasons including availability, security, safety, and risk mitigation.
“If we look at large factories, the whole production would be at risk if something happened with the connectivity or availability of a remote central data center. These systems are mission-critical and need to run even stand-alone to ensure non-stop operation. These local data centers can, however, be managed by the enterprise itself or by a service provider or operator as a service. We see both models in practice depending on the capabilities and strategy of the enterprise customer.”
So, management as a service may still be an option for enterprises seeking to bring their colocated assets closer, particularly to reduce latency. Local breakouts may still be required for telcos to implement select customer-focused functions. But it’s looking more and more like those functions may be sold to the customer like cloud-based services rather than like miniaturized colo facilities, at least as we enter this latest period of experimentation and uncertainty in the development of 5G.