Telcos began using Network Automation in 1891 when they first deployed electronic switches. Today, we can apply AI and Machine Learning to automate much of our Software-Defined Networks. Next, AI will be unleashed to manage business objectives like reducing cost and maximizing revenue. Join Telecom Council’s ComTech Forum members to review the emerging solutions, startups, and opportunities in the next phase of Network Automation.
On an unusually rainy May Thursday, the Telecom Council held our most recent open meeting on the subject of Network Automation. The general thrust of the presentations is that with the trends of virtualization / NFV, Containerization, SDN, 5G densification and service agility, the job of managing and orchestrating the network is growing well beyond the scope that humans can handle. We are no longer being asked to sit in a NOC, and watch for “red flags” from a static network designed into hardware, but rather we must watch a changing, dynamic network, react to demand, outages, provisioning, and new service requests in real time. This many decisions made at this incredible pace requires Network Automation.
Our various speakers explained how Network Automation, at first, can be a simple set of rules programmed into the system by savvy engineers. Then the system can apply these rules, conditions, and heuristics to execute the prescribed reaction. This is an acceptable level of Network Automation to overcome the challenges of scale and speed. But it does not unlock the versatility of telecom networks running, essentially, as software.
To truly unlock the benefits of SDN, we must design and deploy networks managed by AI. Software that can learn, test, and improve the performance as well as, or better, than a human engineer. Of course, engineers will still need to program the objectives, constraints, and limits of the AI. Within those constraints, the AI will optimize towards the objectives in ways that the programmers may not even have conceived, or even that they cannot understand ex post facto.
An interesting analogy kept coming up during or meeting, likening Network Automation to the hot topic of self-driving cars. Similarly, an AI needs to start by being trained, tested, and iteratively improved until it can match a human. Then it can be “set free” to experiment and learn more, eventually surpassing the human and unlocking efficiency gains.
Some factors presented by our great speakers:
Data Feeds AI – AI is trained by feeding it data. More data = faster learning. And telecom networks are rich in data.
Time Crunch – Once an AI is fed data, it can run test simulations on that data, and iterate. This allows it to test in a sandbox. But by allowing multiple computers to test in multiple sandboxes, at a faster clock-rate than real time, allows AI to “crunch time” and test days of simulation time in just hours.
Close the Loop – Automated systems and AI in telecoms will be best when they are not just about “better network” but about “better business results”. As such, telecoms that “close the loop” and include sales, costs, and customer experience numbers into the AI will not just be testing and optimizing the network, but optimizing the business. Later gen Network Automation needs to close the loop with the business results.
Standards – Standards will play an early role in accelerating Network Automation. And learning AI might develop solutions outside of standards, but could also be programmed to stay within the constraints of a standard.
Overall, our meeting enjoyed a number of startup presentations on what could be done for Network Automation right now, combined with a good amount of “blue sky” thinking about the future of Network Automation might bring. Thank you to our host, Nokia, and to all of our speakers. Members can view presentations from this and previous meetings in our Member Library.