Sreekanth S S, Senior Principal Technology Architect, Infosys
Though artificial intelligence (AI) has become a household name, bringing AI to reality is a puzzle which is not yet solved completely. Luckily, the advent of efficient machine learning (ML) algorithms, faster processors,and insightful data collection methods has accelerated the possibility of practical AI in recent time. Telecom is predicted to be one of the early and major adopter of AI in the coming decade because of its need for predictability, faster problem resolution, and elastic performance. The estimated spend for AI-based automation in Telecom is 2–3 BUSD in the next 3 years. One of the key enabler trends in Telecom which brings AI close to reality is software defined networking (SDN) and the associated technology,network function virtualization (NFV).
Endless Possibilities of AI in Networks
AIhas endless possibilities in network with very interesting use cases.
a) Predictive network operations– In a telecom network, network device unavailability is highly catastrophic as it can impact a single customer or a whole metro itself depending upon its location in the network. The legacy network fault correction is reactive, meaning, it is post event. AI can analyze the historic fault patterns and learn to predict the problem based on leading symptoms.
b) Self-healing and learning – AI can monitor and learn human actions. This will help to come up with optimal corrective action over a period of time and also to automate the same.
c) Correlating the customer behavior to network – The quality of telecom services has closed linkage to the corresponding network resource state at the time of service consumption. Therefore, correlating the service usage pattern of a customer to the corresponding multiple network nodes will provide better insights for customer experience improvement. AI can easily achieve this.
There are lot more use cases AI can bring to telecom. The above-mentioned ones are only the tip of the iceberg and are closest to realization.
How SDN and NFV can Power AI
Thinking of why some of these use cases are not very much implemented leads to some of the prerequisite capability for AI. This is where SDN can step in as a powerful enabler.
a) Centralized network brain – In the legacy network, the control plane, which is the so-called brain of a network is distributed in each node. Because of this, the critical real-time information of the network is also distributed. For an AI system to get this information, it has to traverse through the distributed network nodes. SDN, to a great extent solves this problem. SDN centralizes the network control plane and simplifies the multilayered network architecture.
b) Granular data – Data is the key for the accuracy of any AI system. NFV allows you to get granular data up to the functional level. Metrics and performance data can be collected through virtual function managers and orchestrators and can be easily correlated with services running on the network.
c) Easier data access for the AI to consume – SDN/NFV also simplifies the data access. Unlike the legacy method of getting node level data through network management protocols, the virtual function managers and orchestrators expose data through representational state transfer application program interface (REST API) and standard interfaces. These interfaces facilitate easier real-time data ingestion to the data management layer of AI systems.
d) Flexible network actions – While a lot has been talked on data management, it is also important to consider the flexibility of network to handle intelligent actions. SDN and NFV orchestration expose APIs which makes network visible to the upper layers as service (NaaS). AI systems can consume these APIs and feedback automated actions into the network quite easily and in real-time.
Putting together into an example, imagine that there is a network fault which is potential to make your Internet connection break. In the current world, typically, this condition would be known only after fault has occurred. The best case would be that the network monitoring system catches fault and triggers an action either manually or automatically. By then,consumers had already experienced the service disruption. In an AI world, the fault will be predicted based on already learned patterns and leading symptoms. Once the fault symptom is identified as potentially impacting Internet connection, proactive actions are automatically initiated and the network is healed for the condition before the fault happens. For consumers, it is an uninterrupted service experience.
It is clear that to harness the full advantage of AI for telecom the underlying network infrastructure needs to be enabled and this is what SDN/NFV promises. It is rightful that SDN/NFV and AI are getting adopted almost at the same time such that one is complimenting the other. And, together they are powerful enough to create a realistic and intelligent network.