Need to know Edge: No one is buying so no one is building (except China, Amazon, Google, Microsoft & data centers)

In 2019, I thought Edge would be in the telcos

Verizon, DT, Telefonica and other telcos were excited about building Edge servers for low latency. But none of them are installing many outside of China. (So far). Verizon has postponed indefinitely its planned rollout of a thousand Edge servers. Despite brave talk, its “Edge” is mostly press releases and not low latency.

But once the automakers gave up on controlling cars from outside servers, they couldn’t find customers willing to pay. Connected cars are ubiquitous and now come with 5G. But you don’t need low latency if the auto does the driving.

Video doesn’t care. So far, neither does gaming. Reading gas meters and the other remote IoT apps don’t require speed. The 500 billion home IoT almost all choose Wi-Fi & Bluetooth.

No major application needs “ultra-low” latency except VR and security apps. The only large customers I know which signed up were national security organizations. Facebook’s Oculus VR rig is finally selling but is designed to avoid latency limits.

Cloud outfits like Equinix realized a few years ago servers and space were a limited business

“What we’re focused on is not selling space and power, like our focus is to enable globally distributed hybrid multi-cloud infrastructure”

As the telcos learn the business, they are going very slowly on investment.

Hybrid clouds, not Edge servers

Edge clouds alone don’t cut it. Apps need to be tightly coupled with the primary cloud, where storage and processing is much cheaper. Many also need to work with the servers on company sites.

Hybrid clouds are extremely complicated, requiring literally thousands of engineers. Telcos discovered they don’t have the skills and staff to support hybrid. Even attempting to do so is expensive.

The fundamental error here was assuming servers/Edge clouds were simple. Applications and developers expect a rich environment. Kubernetes and Docker are just the start. Take a look at Amazon or Microsoft documentation or any of the college textbooks.

Almost all telcos are giving up on building their own sophisticated cloud. Instead, they are in effect renting space and connectivity to Amazon, Microsoft, and Google. They presumably also get a sales commission when they bring the hyperscalers customers.

Providing space and power to the giants means much smaller profit but saves the enormous capex and especially opex/programming expense.

Fiber, not Mobile Edge Computing

The large potential customers are almost all corporations with robust fiber connectivity. Their endpoints are not scattered consumers who are mobile but large permanent buildings.

Edge plans and 5G networks were planned together. But fiber is faster: generally under 20 ms, while Verizon specifies its 5G as 30 ms.

Fiber is almost an order of magnitude cheaper per bit. That matters to the largest potential customers, who need to transmit a great deal of data.

Data centers have better de facto latency

Lumen/Level 3 can reach 90% of large corporations and data centers in 5 ms. Telcos originally spoke of < 10 ms latency for 5G. Vendors and politicians touted 1 ms. 1 ms is a fantasy outside the lab. Almost no actual 5G is close to 10 ms.

Data centers usually have high speed, low cost connectivity to the clouds. Hyperscalers usually have put their Edge, such as Amazon Wavelength, at regional data centers.

From data center to wireless phones is also generally fast. The telcos have sped up their network from the interconnection to the outside world to their phone customers. So even mobile phones and the data center generally is pretty fast.

Verizon’s slow “Edge” may be typical

Verizon planned to build 1,000+ server nodes, generally at local aggregation points or centralized RANs. The original thought was to go to each tower but that was rejected as too costly.

With 1.000 nodes, they expected to be not much more than 10 ms from your phone. That design was replaced with a much more limited network. Boston, for example, is getting a single node.

From the phone to the node are numerous routers and switches slowing things down. Verizon now specifies 20-35 ms. Average US latency, mostly 4G, is 31-34 ms (Open Signal).

Deutsche Telekom and Korea Telecom, also strong proponents of Edge, appear to be similar. Neither has identified much more than a dozen locations.

China is going big at government direction. Telcos are building large clouds

Duan Xiaodong of China Mobile calls it a “Computer power network.” Unlike the West, the giant Chinese carriers are building massive central clouds as well. They will compete with Alibaba and Huawei, whose clouds are growing even faster than Amazon or Google.

Minister Miao Wei in 2016 set the goal of 25 ms latency to 90% of the country in 2025. They will have 5G to 90% in 2023, with 2,000,000. I haven’t seen evidence of the construction of the necessary regional servers, but the Chinese carriers usually deliver what the state requires.

The Chinese networks and cloud are likely to conform to ITU Y.2501 on Computing Power network framework.and architecture.

But that’s another article

Posted in H and tagged , .

Leave a Reply

Your email address will not be published.