Crucial question: Where goes Edge. 5 ms, 15 ms, 30 ms,50 ms, further

5G network latency delays often are doubled if the telco doesn’t have a well-designed Edge Cloud. I discovered I couldn’t do a good job on 5G Strategies if I didn’t incorporate Edge Networks.

Verizon estimates 30 ms latency on its 5G network. It is installing Edge Networks that will bring latency down to about 15 ms. Vicki Palmer intends to turn on the Verizon Edge in 2019.

The biggest design question about Edge is where will it be?

Edge clouds are a combination of storage and computing power that can be placed anywhere in the network.

For the best performance, you would put many servers near the towers, which is expensive. BT estimates 1,000 sites, with latency probably 15-20 ms.

Vodafone instead is planning to deploy in the regions. That only would require 64 sites but the latency would go up to 20-30 ms. For 5G Strategies, I’m looking at the tradeoffs.

Why Clouds? Clouds save companies a great deal of money by managing a shared data centre. Amazon has built a remarkable business with the AWS cloud. Google, Microsoft, IBM, and now Baidu and Huawei are spending $billions for huge facilities around the world.

It takes time to get your data to and from the cloud. For many applications, a 50-150 ms delay is not a problem. But some applications, as as streaming virtual reality, that delay is unacceptable. VR users get sick if the delay is more than about 20 ms.

The solution is to bring the servers closer to the user.  1. I developed this framework to organize the analysis.

Customer premise

Modest

This this could be as modest as as a single server computer or even home units from Google or Amazon. For big homes or modest businesses, it can be more developed.

Substantial

Large businesses and factories often want to keep their data in house. The latency is lower because the time to the telco is eliminated.

Amazon, Google, and Microsoft have very ambitious programs for a “distributed cloud.” They put a server or server cluster on the customer premise which they manage. Applications and files are divided between the on-premise server and the big iron in the sky.

Managing a cloud is challenging. Most organizations – even telcos -would be wise to let the specialists take over. Some telcos believe they have the expertise to do on-premise cloud management and expect major revenue.

This analysis is not covering customer premise servers but I might analyze them in the future.

Edge Clouds inside the telco network

Deutsche Telekom in early 2019 was the first telco to offer customers access to an Edge Cloud from inside its network. Working with MobiledgeX, it has installed a Level 4 Edge Cloud, 20-30 ms away from the customer.

Verizon is committed to deploying in 2019. SK in Korea, AT&T, and the Chinese are well along but haven’t announced. All the major telcos are in some kind of trials.

Every network is different. I find the framework above useful, with four lavel of cloud:

Level 1 Edge Clouds go right at the cells or the towers. They may cost too much to deploy. No one has built any yet, although the Chinese might. The “air latency” from the phone to the tower is ~10 ms. With some delay in the server, Level 1 clouds will have 10-15 ms latency.

Level 2 is one to three hops back. Latency should be 15-20 ms. The servers could be at the C-RAN that controls multiple cells or at the exchange. British Telecom estimates about 1,000 servers would be required.

Level 3 sites are still further back, perhaps in the regions. Deutsche Telekom has a system now accepting customers. Vodafone believes a similar system for Britain would require 60-70 servers. Probably 20-30 ms.

Level 4 are at the other extreme edge, where the telco network connects to the Internet and other services. Telefonica, Verizon, and others are simplifying and speeding their networks, bringing the two edges much closer. The faster backhaul/transit often reduces latency more than the switch from 4G to 5G. On the new networks, even a few servers can offer a service with less than 30 ms delay.

Level 5 is what we have today, on the other side of the peering point. My Netflix and everything else runs fine. Maybe that’s enough. Akamai, Amazon, and many others bring the bits to the telco, reducing the delays.

LTELevel 530-75+msToday
5G or LTELevel 425-50+ msGood transport
5G with servers in telco coreLevel 320-25 msDeploying in DT and
China
5G with few hopsLevel 210-20 msFirst units getting to
the field.
5G with 1 ms air
latency
Level 15-12 msNot expected out of
the labs for years.

  1. Close refers both to the physical distance and the delay factors in the network. Data from your phone has to go through routers and switches to get to the servers. A telco usually has 4 to 10 steps to go through the network called hops. Light moves fast, so usually the number of hops is the bigger factor.

Improvements welcome