As telecoms operators continue to transform their networks and businesses, the industry is in the midst of the greatest change since the introduction of IP networks more than 20 years ago. The era of manually-managed, function-specific network equipment is coming to an end as operators deploy commoditized hardware, the function of which is software-defined and responsive to the demands placed on the network, especially when bolstered by the likes of artificial intelligence (AI), namely machine learning. This presents operators with exciting new opportunities to significantly reduce their costs, making new capacity investment sustainable, while enabling new service revenues thanks to the flexibility of software-defined networks.
Without a doubt, greater value will be ultimately generated from the new service arena but, more immediately, operators will save capital expenditure on hardware and start to make operational expenditure gains from increased automation as they move away from costly and slow manual processes. This transformation represents a step change for the telecoms industry and means that operators can escape from being confined to the role of network providers in the digital value chain. However, in the shift to virtualized infrastructure and self-organizing networks, it’s important that the attributes of the telecoms industry are not lost. Telecoms has always prided itself on offering carrier-grade, robust, high-availability services and that network resilience needs to continue in the virtual era if users are to continue to receive the experiences they want.
This will be a different challenge with virtualized infrastructure, which changes function continually, as different services create different network demands at different times of day and user demand for capacity continues to grow. Operators, therefore, will face a series of challenges.
Their personnel will need to transition from managing function-specific hardware to managing commodity equipment via software. Much of this management will need to be automated to keep costs down. Of course, network engineering expertise will still be essential, especially during the decade-long period of hybrid operations when physical and virtual networks will work in parallel. The difference will be that this engineering experience will be applied in software terms.
Automated systems will not be able to operate in a vacuum; they will need to be fed continuously with data insights in real time to maximize capacity utilization and ensure services are delivered in the quality expected by users. This will rely on the terabytes of data generated by networks daily, but the data in itself has only limited value. The real advantages come from applying analytics to the vast volume of operator data in order to extract actionable insights. Analytics, therefore, is the key to efficient operations for software-defined operators. The challenge here is to accelerate the analytics process so the valuable insights can be achieved in minimal time. This is so important because the network will be continuously changing functions according to demand and therefore the task is never complete; insights are needed all the time as different services create different demands on the network at different times.
As the concepts of AI and machine learning are becoming more familiar in the consumer market, with the introduction of services such as Siri and Alexa, it’s clear that the technologies have applications for telecoms operators. AI can be used to accelerate the analytics process by identifying repeated behaviors, thereby enabling the automated system to take action based on previously successful decisions. Machine learning can be utilized here when such a pattern is identified to enable a learned response to be utilized.
A good example is that of a snow day on the U.S. east coast. This means children stay home from school and consequently residential neighborhoods experience huge spikes in video downloads at a time when the network would normally be serving a small number of home workers. Data from the network would identify this increase in demand but artificial intelligence could then be applied to that and augmented with weather information to enable the network to understand that video demand is likely to continue at a higher level for the rest of the day. Machine learning would then come in to provide the network with a learned response for it to configure itself, perhaps by instantiating more video content servers, to serve the demands of the day.
There will be many different types of service profiles created that effectively become a library of learned responses for automated network operations. These have the potential to radically shorten and accelerate the decision-making process, enabling the reality of the flexible, dynamic network at a sustainable cost of operations.
Machine-learnt knowledge can, therefore, be deployed to achieve optimum route design and configuration of networks, ensuring maximized utilization and minimization of the cost of capacity overbuild but the story doesn’t end here. Machine learning and AI can be utilized to enable predictive analytics. In the snow day scenario, the system could learn that if weather data indicates snow is falling and the day is during the school term, children will stay home and stream video. A machine-learnt response can then be made so the issue of lots of children suddenly using the network is addressed before it becomes a service-affecting issue.
This type of approach can be used in so many different ways across the network, utilizing insights from location data and many other systems to create decisions in advance that solve network problems before they're detectable to the user. This could involve ensuring capacity is available to serve crowds at a sports event or recognizing that a hardware failure means alternative capacity needs to be provisioned before the network comes under strain.
The current challenges are that machine learning and artificial intelligence are relatively new concepts and it’s vital to ensure that machine-learnt information is accurate. If operators are to rely on machine learning, this is critical. Fortunately, if the machines are taught well, they will not make errors like humans inevitably do. This heightened reliability means data insights even have the potential to be utilized and shared outside the operator.
Initially, we will see network insights being federated across other business units within operators to feed marketing and customer care as well as network management, but the ultimate prize is to federate this data across the wider partner ecosystem, communicating valuable data insights to third party organizations and generating revenue from that.
~ Written by George Malim. George is a freelance journalist who covers the telecoms and internet markets.