DENVER -- Big 5G Event -- If a 5G network's edge can be distributed, where should it be? That issue came up several times at this week's Big 5G Event in Denver as vendors and operators discussed how to turn advanced 5G applications into reality.

In addition to being designed for RAN latency of less than 1ms, 5G will allow for virtualization and distribution of the systems behind the RAN, including baseband units remote from radio head and computing moved away from centralized clouds. Exactly where those elements should reside will depend on distance and cost, and no yet one knows where the edge should be, Heavy Reading analyst Sterling Perrin said. There may even be more than one.

The original concept for mobile edge computing (MEC) was to place servers for application processing in a macro cell site, which would also put them close to small cells. But in the past year, carriers have balked at the cost of that approach, Perrin said. Now they are looking at placing it in their central offices (COs), which would mean deploying to just a fraction of the number of sites. And there are other options, including processing at the customer site and partnering with operators of local data centers.

Depending on how many low-latency services carriers can sell, any or all of those options might pay for themselves -- but the operators don't yet know what services will take off. "You have to spend a lot of money to make a bet, and it's too early to make a bet right now," Perrin said.

It's not even clear what all the options are for where to locate the edge, Perrin said. This could be the next question for the industry to tackle. "I think the menu is going to be more clear in a year," he said.

Infrastructure providers such as Zayo Group stand to gain a lot by hosting edge infrastructure in their facilities. On Wednesday, Zayo agreed to a $14.3 billion buyout by two investment firms. The company owns 12.2 million miles of dark fiber in North America, the UK, France and Ireland, and offers managed services.

Mobile operators need to decide where to place the 5G edge as they adapt to demanding new applications such as connected autonomous cars, said Brian Daniels, Zayo's senior vice president, strategic networks, Z5G. He spoke at the event on Tuesday.

MEC all the way out at macro sites is more than just a monetary challenge, Daniels said. Most existing cell sites aren't equipped to host what is essentially a small data center, because they lack features such as redundant power.

Further complicating the problem of where to put the edge is distance, which affects latency even if the link between the elements is fiber, he said. Applications that work downtown might not work in surrounding areas or suburbs without edge infrastructure closer to the site.

Applications will drive service providers' choices about where to locate the edge, but it's not clear which ones, Daniels said. "Nobody has a concrete idea of what's going to put the dollars around this," he said.

AT&T plans to use several edge computing approaches to cut what is roughly a 100-millisecond roundtrip latency between customers and a central cloud, said Jeff Shafer, associate vice president of edge solutions and portfolio transformation.

Computing in the metro network, with no need to send data to the central cloud, can cut the roundtrip latency to 20ms, he said. Moving data processing all the way out to the customer edge, in servers owned by the customer, carrier, or local data center company, can slash that to 10ms or less, Shafer said.

For a test of drone-tracking technology from Israeli startup Vorpal, AT&T has set up servers at the network edge to do real-time processing of data coming from Vorpal sensors. The passive sensors, connected to the mobile network, use the drone's radio signals to track them. From the mobile network, that data goes to the servers at the edge of the network. This provides the latency required for real-time tracking while not requiring server rollouts to the test site. Servers can be deployed and scaled more quickly at the network edge, Shafer said.

Why this matters
For 5G networks to deliver the digital transformation that enterprises have been told to expect, changes need to take place across transport networks and computing infrastructure as well as on the RAN. Large commercial deployments of advanced applications may have to wait until the range of choices for how to make those changes is broken down. That effort will require cooperation among wireless operators, service providers, vendors and potential customers.

— Stephen Lawson, special to Light Reading. Follow him on Twitter @sdlawsonmedia.