Contributed by Anil Rao, Principal Analyst, Analysys Mason. 

Communications service providers (CSPs) are introducing network function virtualization (NFV), cloud-native computing (CNC) and software-defined networking (SDN) technologies into 5G networks, even as they continue to operate legacy physical networks. This mix of legacy and new networks increases the operational complexity that cannot be managed using traditional operational approaches. The legacy operational systems and processes were predominantly designed as siloed software solutions for various individual network domains and required significant manual effort and intervention. This approach is not economically and operationally viable for complex 5G networks and services. Autonomous networks will be critical to make legacy operations lean and to make 5G successful.

However, the journey to fully autonomous networks will be gradual; CSPs need to put strategies in place to carefully manage the transition. The TM Forum model for autonomous networks uses levels (0–5) to assess the maturity based on a set of conditions.  According to a research conducted Analysys Mason,  most CSPs are at level 2 (partially autonomous networks with limited autonomous capabilities), and some advanced CSPs are operating some network domains at level 3 (conditionally autonomous networks: intent-based automation based on real-time network changes).

Analysys Mason’s reference framework for autonomous networks

There is a range of network and service automation visions and projects set by CSPs and vendors, as well as numerous industry initiatives by standard bodies and groups including TMF Autonomous Networks Project, ETSI Zero-touch Network and Service Management (ZSM) and ONAP, all of which are striving towards enabling a high level of end-to-end automation (level 4 or 5). A common understanding and a high-level framework for enabling autonomous operations is emerging from these efforts. Figure 2 illustrates Analysys Mason’s reference framework for autonomous networks.

To achieve the vision of autonomous networks, CSPs need to adopt an intent-driven automation approach. CSPs must first define the business intent; this could be an enterprise wanting to automate their WAN or the CSP wanting to fully automate its mobile broadband operations. The business intent is then translated into service intent (encapsulating the service design and service orchestration processes) and network intent (network resource and service orchestration), supported by assurance systems to monitor the state of the network and services. Together, the service design and orchestration, network orchestration and automated assurance systems, supported by AI/ML capabilities, enable an intent-driven, fully closed-loop autonomous network. (Illustrated in Figure 1)

Figure 1: Framework for autonomous networks

Source: Analysys Mason, 2020

CSPs should build a hierarchical autonomous network architecture using a stepwise approach

CSPs need to implement a hierarchical cross-domain automation architecture that collapses network siloes, simplifies network domains, and transforms the underlying network into a platform by abstracting the underlying complexity. The use of industry-standard open APIs allows CSPs to achieve this by enabling adjunct applications such as SDN controllers and other OSS applications to access the network in a standardised way. A unified AI/ML and network analytics platform can further accelerate the journey to level-5 multi-domain autonomous networks.

Figure 2: Main building blocks of autonomous networks

Source: Analysys Mason, 2020

CSPs should build autonomous domains in a stepwise fashion with domain-level local abstraction and orchestration for intra-domain closed-loop automation and self-sufficient operations (self-configuration, scaling, healing and optimization) over hybrid physical and cloud resources.

• These multiple autonomous domains should then be integrated into an end-to-end cross-domain service automation layer using open northbound APIs for service automation. The cross-domain orchestrator should stitch together the different domains and automate the resource selection, configuration and optimization for the service (for example for creating end-to-end network slices across the core, transport, edge and RAN).
• Cross-domain service automation capabilities should be exposed to a business and ecosystem automation layer for agile, on-demand service delivery and customer and partner management.
• Advanced analytics with AI/ML will be pervasive in this architecture; there will be several instances running across these layers and embedded in the network nodes and elements. CSPs should create centralized data lakes and a common set of AI/ML tools, processes and governance to further increase efficiencies.

AI/ML will be essential in all layers of autonomous networks

CSPs will need advanced AI/analytics capabilities in all networks at level 3 and above. Such capabilities can aid self-learning from fast-changing environments, provide accurate predictions of potential network problems and anticipate trends and ensure the continuous improvement and adaptation of the rules that govern autonomous operations. Our interviews revealed that advanced CSPs typically start with piecemeal, tactical implementations of AI/ML with various tools, methodologies and datasets that are specific to a network element, domain, use case or business problem. However, as they expand the number and scope of these individual implementations and their sub-components over time, they plan to join them together to create end-to-end operations with centralized data lakes and common tools, models, processes, and governance. This suggests that autonomous networks should be developed with a layered AI/ML and analytics approach that is aligned with the hierarchical architecture discussed in the previous section. The three layers in this approach are described below (Figure 3).

• At the network infrastructure level, intelligent infrastructure with near real-time data collection and embedded AI/ML inference capabilities executes local closed-loop management scenarios based on domain-level or cross-domain level frameworks and algorithms.
• Domain-level data collection, intelligence, analytics and knowledge management will support variable degrees of automated decision making (levels 2–5) inside each autonomous domain. This layer should monitor, collect and analyse live data streams from intelligent infrastructure within a single domain, provide domain-specific insights and predictions and trigger actions based on policies and scenarios.
• CSPs will also need an overarching unified and centralized AI platform to enable cross-domain and complete closed-loop service automation. Such a platform is responsible for aggregating and federating data sources, generating AI/ML models and supporting cross-domain automation with end-to-end service insights, KPIs, predictions and actions.

Figure 3: The three-layered AI/analytics approach for autonomous networks

Source: Analysys Mason, 2020

Preparing the organization for autonomous networks

The largest obstacle to making progress towards autonomous networks is organizational readiness. Highly skilled personnel in network engineering and operations departments have developed a rich body of knowledge that is often undocumented. This expertise is typically formed from years of performing manual procedures, so engineers tend to be more practical than strategic and are deeply resistant to change. CSPs can therefore find it extremely difficult to impose the new practices required to achieve the goal of autonomous networks. As such, CSPs need to devise a multi-pronged strategy to prepare their organizations for autonomous networks (Figure 4).

Figure 4: How CSPs can prepare their organizations for autonomous networks

Source: Analysys Mason, 2020

Conclusion and recommendations

CSPs are in the midst of a long journey towards autonomous networks. Analysys Mason’s research has established that most CSPs are at level 2 in TM Forum’s autonomous networks framework, though some advanced CSPs are operating some domains at level 3. The CSPs that we spoke to have built the technology foundations to either make the transition to level 3 or achieve a complete level-3 autonomous network.

To accelerate the journey towards autonomous networks, Analysys Mason makes the following recommendations for CSPs.

• CSPs should have a roadmap for achieving autonomous networks. As networks become more complex and dynamic, CSPs will find it increasingly difficult and expensive to manage them. In the current mode of operations, business processes such as network design, planning, fulfilment, and assurance are highly disjointed, and often require manual interventions and inter-departmental handovers. This model is not sustainable in the emerging telco cloud network environment. CSPs will need a new operations model that is based on a vision for autonomous networks.
• CSPs should pick a starting point and a preferred automation approach that can quickly yield results. Automation approaches vary based on each CSP’s market situation and business strategy. As a first step, CSPs should consider a domain-based automation approach with enabling technologies such as ML/AI, open APIs, network abstractions and orchestration to fully automate the domain. CSPs should also implement a horizontal cross-domain strategy for end-to-end automation.
• CSPs should transform their organizations for autonomous networks. Organizational transformation is much harder to achieve than technological transformation, but it is critical for success. CSPs should reskill their existing workforces alongside hiring new talent with the requisite DevOps and software engineering skills. CSPs should also consider seeking support and should adopt best practices from external vendors to ease and accelerate the journey of organizational change.

About the author

Anil Rao (Principal Analyst) is the lead analyst for the Automated Assurance and Service Design and Orchestration research programmes, covering a broad range of topics on the existing and new-age operational systems that will power operators’ digital transformations. His main areas of focus include service creation, provisioning and service operations in NFV/SDN-based networks, 5G, IoT and edge clouds; the use of analytics, ML and AI to increase operations efficiency and agility; and the broader imperatives around operations automation and zero touch networks. In addition to producing both quantitative and qualitative research for both programmes, Anil also works with clients on a range of consulting engagements such as strategy assessment and advisory, market sizing, competitive analysis and market positioning, and marketing support through thought leadership collateral.