Monday, January 4, 2021

The telco multi core

TobiasD / Pixabay

There is something that has been irking me for the last few months: everyone in telco seems to carry on thinking that they will continue have a single omnipotent centralized core network. Even though variations between workloads (voice vs browsing vs video vs gaming vs AR vs AI vs IoT...) continue to amplify and the business models (owned, and operated, IaaS, SaaS, PaaS...) increasingly require separate command and control.

The answer seems to be that slicing will magically solve everything. I fail to understand how slicing can accommodate diverging simultaneous needs from the same infrastructure without overprovisioning but that's a question for another time.

What troubles me most, is that networks have dealt with separate cores for a long time. In many cases, because of IoT or B2B business units who could not afford the timelines and costs of adapting the centralized core, or because, simply the network authority wanted to separate consumer traffic from enterprises. In other cases, you have network sharing and multi-operator core networks (MOCN) that have emerged as viable solution to segregate and manage traffic in a logical network.

I am not an engineer or a scientist, but it feels like the most advancement in processing in the last years is due to parallelization or specialization, and I don't see silicon vendors building bigger CPUs, but rather orchestrating as many CPUs on the same board as possible to manage concurrent, yet different workloads. This analogy has also seen the emergence of specialized processing units such as GPU or TPUs for specific workloads, in specific circumstances...

Now that most cloud providers and many telco vendors have proven the compatibility of their core network (at least the control plane) with cloud infrastructure and networks, I don't understand why telco standards and industry still feel that 5G will have THE core network to evolve to, and that, when, it will be 5G, when it will be standalone, when it will support slicing, when it will have a platform to recognize, identify, reserve, network resources, when it will be able to create dynamic slices on demand... all will be solved.

I feel that many of these issues have been resolved yet? Slicing is just a new iteration of tunneling, VPN, packet tagging, traffic shaping that are today prevalent in many networks. Cloud providers have effectively solved most of these challenges within their networks already so why are telcos trying to reinvent the wheel? 

Wishing a single, unique, centralized core is not necessarily going to make it so. Other telcos, cloud providers, soon industry verticals, governments, IT vendors will have their core. Thinking that the telco single core architecture will be able to manage all workloads and use cases and verticals simultaneously in a 5G world seems too much like magical thinking.

If you're a telco, you might not like it but you better plan for a multi core network, because others will be soon, whether you want it or not. Chances are there are already premises in the third party caches and edge infrastructure being deployed in your networks.

You might want to start thinking in terms of core per service types, like voice, unicast TV, general browsing, low latency IoT, high compute applications, Edge... and per business model like retail consumer, retail enterprise, wholesale telco, wholesale cloud, IaaS, PaaS...

No comments: