Last week’s Zero Touch Automation Congress in Madrid had a heavy focus on edge computing, and the UK’s BT was one of the operators providing details of its plans.
The company said it aims extend its cloud platform to about 100 metro locations to support new 5G and converged network applications. The company has previously said it would not require more than 10 edge data centers to support services such as high quality video for the whole UK.
But that was related to classic telco services such as content delivery, which do not require a highly distributed edge to improve quality of experience. By contrast, many of the edge-enabled services with high growth potential, in industrial environments, will need far more locations and even an on-premises edge data center, to deliver the hoped-for benefits of low latency, reduced data transmission costs, and improved security and control.
Maria Cuevas, BT’s head of mobile core networks research, told conference attendees that BT’s core network functions are currently hosted in 5-10 major exchanges across the UK. Under the telco’s Network Cloud program, it will increase this number to about 100 metro exchanges, which will be equipped to host enterprise customers’ network applications and so extend the edge computing business model for BT, from applications for its own retail customers, to an enterprise and wholesale play.
A future phase is likely to see an even more dispersed edge, with edge cloud capabilities deployed in some of BT’s central offices and local exchanges. There are about 1,200 such centers across the UK, providing the first point of aggregation in the BT network.
Cuevas said: “We need the flexibility to be able to deploy different functions in different parts of the network based on the needs of the service. Ultra-low latency means bringing user plane functions closer to the edge of the network. That dynamic flexibility will underpin a lot of 5G investment and opportunities for the future.” BT is reportedly targeting latency reduction from 30ms to 20ms on average, in the first year of deployment of 5G by the firm’s mobile subsidiary EE. The medium term target is below 10ms, which will enable a new set of 5G use cases including dynamic robotics and drone services.
“We’ll start with the first level of distribution to roughly 100 nodes across the network and will look to further expand according to the needs of different services,” she said, though no timescales were disclosed.
That figure of 100 for the UK sounds modest compared to the thousands involved in the industrial edge, but for one UK operator, even 100 is too many. Vodafone UK’s CTO, Scott Petty, thinks 100 is “vendor wishful thinking … The use cases don’t exist for why we would want to do that”. Instead, he thinks Vodafone UK would need about eight data centers to guarantee latency of 15-20ms in the UK, and even to support new ultra-low latency or industrial use cases, he does not see the need for more than 64 edge nodes.
“The power consumption, security concerns and energy costs would outweigh the economic benefits,” he said.
He also thinks telco edge applications will not appear before about 2023 because of the need to “figure out how to put in the right telemetry tools and operational capabilities”, as he told Light Reading.
Roadmaps like BT’s still highlight a gulf between telco thinking and that of cloud or industrial edge companies. While BT is not fully convinced it will need to move beyond 100 locations any time soon, Google has said its new cloud gaming platform, Stadia, would require several hundred nodes per market (far more in the USA). It is not clear how well telco sites will align to the needs of cloud and industrial partners.
And though operators have hoped that their locations will be attractive to the highly centralized cloud companies, resulting in new revenue sharing, Google says it already has 7,500 edge nodes and those will be sufficient to support Stadia in the markets where it will be available.
Rather than partnering with telcos, some cloud companies are making their platforms more dispersed by installing edge computing resources in data centers built by Equinix or other wholesalers.