They are regulated, they are critical infrastructure, they require a level of engineering and control that goes beyond traditional IT. This has often been the reason why some technologies and vendors haven't been that successful in that space, despite having stellar records in other equally (more?) demanding industries such as energy, finance, space, defence...
Being Telco grade, when I cut my teeth as a telco supplier, meant high availability (5x9's), scalability and performance (100's of millions of simultaneous streams, connections, calls, ...), security, achieved with multiple vertical and horizontal redundancies, and deployed of highly specialized appliances.
Along comes the Cloud, with its fancy economics, underpinned by separation of hardware and software, virtualization, then decomposition, then disaggregation of software elements into microservices. Add to it some control / user plane separation, centralized control, management, configuration, deployment, roll out, scalability rules... a little decentralized telemetry and systematic automation through radical opening of API between layers... That's the recipe for Cloud grade networks.
At the beginning, the Telco-natives looked at these upstarters with a little disdain, "that's good for web traffic. If a request fail, you just retry, it will never be enough for Telco grade...".
Then with some interest "maybe we can use that Cloud stuff for low networking, low compute stuff like databases, inventory management... It's not going to enable real telco grade stuff, but maybe there is some savings".
Then, more seriously "we need to harness the benefits of the cloud for ourselves. We need to build a Telco cloud". This is about the time the seminal white paper on Telco virtualization launched NFV and a flurry of activities to take IT designed cloud fabric (read Openstack) and make it Telco grade (read pay traditional Telco vendors who have never developed or deployed a cloud fabric at scale and make proprietary branches of an open source project hardened with memorable features such as DPDK SR-IOV, CPU pinning so that the porting of their proprietary software on hypervisor does not die under the performance SLA...).
Fast forward a few years, orchestration and automation become the latest targets, and a zoo of competing proprietary-turned-open-source projects start to emerge, whereas large communities of traditional telco vendors are invited to contribute charitably time and code on behalf of Telcos for projects that they have no interest in developing or selling.
In the meantime, Cloud grade has grown in coverage, capacity, ecosystem, revenues, use cases, flexibility, availability, scalability... by almost any metrics you can imagine, while reducing costs and prices. Additionally, we are seeing new "cloud native" vendors emerge with Telco products that are very close to the Telco grade ideal in terms of performance, availability, scalability, at a fraction of the cost of the Telco-natives. Telco functions that the Telco-native swore could never find their way to the cloud are being deployed there, for security, connectivity, core networks, even RAN...
I think it is about time that the Telco-natives accept and embrace that it is probably faster, more cost efficient and more scalable to take a Cloud-native function and make it Telco-grade than trying to take the whole legacy Telco network and trying to make it Cloud grade. It doesn't mean to throw away all the legacy investment, but at least to consider sunsetting strategy and cap and grow. Of course, it means also being comfortable with the fact that the current dependencies of traditional Telco vendors might have to be traded for dependencies on hyperscalers, who might, or not become competitors down the line. Not engaging with them, si not going to change that fact. 5G stand alone, Open RAN or MEC are probably good places to start, because they are greenfield. This is where the smart money is these days, as entry strategy into Telco world goes...
No comments:
Post a Comment