Analytics, Machine Learning & AI in Next-Gen OSS/BSS
30 NOVEMBER 2017
Dima Alkin, VP, Service Assurance Solutions, TEOCO
Recently, I was honored to participate in Light Reading’s expert panel on the subject of Machine Learning (ML) and Artificial Intelligence (AI) for NFV/SDN. As we engaged in a lively and sometimes heated debate, one thing was very clear – that telecom service providers are ready to accept a wider role for Machine Learning and AI in their daily operations, and it is now up to OSS providers and internal IT teams to deliver on those expectations.
With the introduction of 5G, the amount of actionable data generated by the network is going to become overwhelming. The industry’s traditional approach towards managing this flood of information has been to throw more people at the problem, and, where applicable, introduce simplistic forms of automation for managing some of the more repeatable and innocuous network functions. This has worked fine up to now, but things are about to reach a tipping point. With the sheer volume and complexity of data that is coming with 5G, this ‘catch-as-catch-can’ mode of operations is no longer scalable.
Previously, significant technology evolutions like 3G and 4G have been introduced in a gradual, incremental mode. 5G is different, in a sense that it is really a culmination of multiple technologies coming together to enable an overarching digital transformation, with many new technologies and architectural innovations being introduced in parallel. NFV/SDN in various parts of the network, edge computing, new radio access technologies, the introduction of fronthaul, and more – all of this is happening together.
All this change is creating the need for us to rethink everything about how networks work and how they’re designed and managed. The use of Wi-Fi hot-spots and other unlicensed spectrum radio access technologies will be leveraged in new ways to provide the required network coverage and speeds that new services demand. Not only will we have hundreds of thousands of traditional small-cells of various kinds being incorporated into practically every nook of city life, we will also have consumer-type devices like Sprint’s Magic Box or the 28Ghz 5G small cell radio that Verizon is experimenting with, that will in effect act as small cells, causing the traditional separation between network equipment and the user device to begin to blur.
Taking the Magic Box concept one step further, soon a subscriber’s handset or an IoT device will itself serve as a relay for other subscribers or “things” on the network; in effect, creating millions of walking small cells. On one hand, we can’t treat all these devices as true Network Elements, using our full ability and toolsets to control and modify them at-will to suit the needs of the network. But at the same time, we can’t ignore the performance and potential malfunction of these devices by treating them as we treat subscriber devices and CPEs today – where their performance and value is oriented entirely towards the individual customer’s needs. We will have to find ways to collect the valuable data from these devices, then aggregate and process it to provide meaningful, actionable insights that can help maintain required network performance, without putting too much dependence upon any individual device. Again, this is just one of the challenges that simply cannot be addressed by current OSS tools and processes. The use of ML and AI is a must for managing this type of complexity, and this is just one instance of many.
Machine Learning and Artificial Intelligence
Artificial intelligence (AI) is classically seen as an autonomous machine that perceives its environment and takes actions that maximize its chance of success at some goal. It resembles human cognition, which can allow it to make judgment calls and responses to a set of tasks.
Machine learning is a subset of AI. It comprises an engine that controls decision-making under a fixed set of circumstances. Machine learning tools become more intelligent as they are fed training data that gives them examples of how to ‘think-through’ challenges and issues and come to the right conclusion.
It’s easy to see how both AI and machine learning could play a part in the evolution of service assurance to match the demands of today’s real-time digital world. Their roles are primarily within automation, where there’s a shift from traditional network monitoring to services-oriented assurance, which takes into consideration the context of services consumed.
Getting there from Here
Machine learning and AI are set to make a huge impact in the way that network health is maximized. But that’s not to say these technological advances will immediately take over the command and control of the network. Or the healing of it when things go wrong. Rather, they will become assistants to Network Operating Center (NOC) and System Operating Center (SOC) staff who are under pressure right now to manage tremendous complexity, change and growth.
As challenges continue to increase and more network management automation is being considered, two areas will come under scrutiny. Firstly, the maturity of the available technologies and solutions will dictate the pace that network operations are augmented with ML and AI. For instance, will the tools provide the NOC and SOC teams with access to actionable, practical insights, not merely descriptive analytics? Will the ML-driven engines be able to make use of all the available data at hand? And can this information be consumed and processed without major data normalization and cleanup requirements?
Secondly, and maybe even more importantly, the maturity of the organization itself, and its appetite for change, will be key to moving to the next step. Organizations with a higher maturity level will more easily be able to effectively consume and act upon the insights delivered by analytics, as well as manage the challenges associated with integrating machine learning and AI into the existing business processes (or modifying those to take full advantage of the newly available technologies). All organizations should look for ways to integrate these new technologies into their operational processes to better enable smarter, more comprehensive automation. It’s not about replacing jobs, it’s about the ability to leverage every resource available to keep up with the tsunami of data that is about to hit.
Advocating for the Customer
The addition of AI and machine learning will drive a higher level of customer experience where SOC and NOC personnel act as consumer advocates, not just network operations staff. These new tools allow a quantum leap in the way the customer is served, offering engineering another perspective on how services are perceived when they are used, rather than monitoring and troubleshooting merely the underlying network resources.
Automation doesn’t replace the knowledge provided by an experienced engineering staff. Instead, it allows them to handle the ever-growing amount of real-time information. This includes all network, IT and services related data; all events, alarms, performance metrics, service usage records and logs – the information that will be increasingly hard to efficiently process, correlate and act on in the current manual or semi-automated fashion.
It’s an important step to take to future-proof your organization, because it will streamline network operations, make insights easier to see and problems easier to fix. The implication is that the expectation consumers have from their CSP – grown from the enjoyment of real-time digital services – can be met. As 5G and other new technologies roll out, consumers will be immersed in a richer service experience without hassles or delays.
As leaders in the Service Assurance space, we recognize the challenges operators are facing. TEOCO has made significant investments in decoupling the analytical capabilities of AI and Machine Learning from the service assurance application layer. This means CSPs can benefit from these insights without changing or replacing their existing systems, allowing them to begin their journey to more advanced automation without disrupting their existing OSS/BSS environments.