With unabated traffic growth continuing to pose cost and complexity challenges for service providers, the roles of automation, artificial intelligence (AI) and machine-learning-type capabilities in transport network infrastructure is consistently gaining more attention. The volume of data on transport networks poses operational challenges, as network planners know well, but so do the increasingly unpredictable traffic patterns and workloads those multilayer, multivendor and multidomain networks must support.

The monetary implications, both capex and opex, quickly compound with the rising adoption of higher-bandwidth and more-latency-sensitive customer services and applications such as 5G, augmented and virtual reality (AR/VR), and the industrial internet of things. In addition, the technologies enabling these new services, as well as the services themselves, demand a new level of operational responsiveness that manual processes no longer can support. As the pace of digital transformation accelerates in telecom environments, this lack of operational responsiveness will sound the competitive death knell for many network operators.

Despite progress on multiple fronts, transforming telecom infrastructure with automated and intelligent operations remains in the nascent stages. In the coming years, innovations in this field will yield important shifts in how network architectures are built, scaled and more effectively leveraged for new service deployment and network monetization.

The topic of software-driven automation in telecom networks is not a new one. Software-defined networking (SDN) has been talked about in the industry for more than 10 years. It has survived the hype cycle long enough to see initial deployments among service providers after revolutionizing data center operations and architectures. What is new, however, is the move among many service providers to rethink their automation strategies and begin employing a more practical approach to extracting value with automation. This practical, building-block approach to automation stands in contrast to the large-scale, all-encompassing orchestration projects that many operators have been mired in over the last several years because of the massive overhaul implications of existing operational paradigms.

Figure 1: Changes in traffic patterns and network complexity drive the need for network automation.

Practical, Application-Based Network Automation

The underlying concept is not revolutionary, and proof abounds in fields as far-ranging as medicine, education reform and urban planning. The basic idea is this: When facing large-scale and highly complex problems, start by taking smaller versions of the larger problem. For service providers, this means identifying specific network problems or use cases and applying incremental, easy-to-use, ready-to-deploy automation tools to help drive business value. This app-based, building-block approach to automation is not meant to replace larger, holistic initiatives but to augment these efforts with quick returns and valuable learnings that can be applied to evolving automation strategies.

Key practical automation applications include

  • Network and service migration – Accelerates modernization and enables significant savings in capex and opex.
  • Automated network discovery and real-time awareness – Enhances overall network performance and improves resource utilization.
  • Closed loop automation – Streamlines operations and enhances network reliability.
  • Autotunable DWDM – Service providers plug dense wavelength division multiplexing (DWDM) optics into aggregation and access nodes, and the system automatically tunes each optical signal to the appropriate wavelength. As a result, a service provider can simplify installation and reduce sparing costs and truck rolls.
  • Automated bandwidth on demand – Enables software activation of service-ready optical capacity in minutes.

Verizon, for example, took a practical approach to automation to address a specific networking challenge – the migration of its time-division multiplexing (TDM) network to a new, simple, Ethernet-based network. To meet this challenge, Verizon employed software-driven virtualization and automation tools that enable simplified, multilayer visualization of service modeling; the ability to run multiple migration scenarios; and the efficient management of existing customer traffic.

Prior to wider-scale implementation, Verizon deployed the technology within a select region, a practical step that enabled validation of the capabilities and associated cost savings while avoiding the “all or nothing” mentality that represents one of the common pitfalls of automation projects. In addition to simplifying and accelerating network and service migration, Verizon was able to realize significant cost savings, recover space and power and improve operational efficiency and network reliability.

In another example, Telia Carrier, an international service provider with one of the world’s most extensive fiber backbones, applied automation technologies in a multidomain portion of its European backbone to address a specific use case – the understanding of the real-time state of residual margin. Residual margin represents the most useful measure of received signal quality in optical networks and determines how much room there is for the signal to degrade without impacting error-free operation.

Residual margin is impacted by the optical signal-to-noise ratio (OSNR), linear impairments and nonlinear impairments. Measured in decibels (dB), residual margin indicates by how many dB OSNR can be reduced, if linear and nonlinear impairments remain constant, until post-FEC (forward error correction) errors occur. Accurately assessing the residual margin requires accurately determining the impact of all three. Powered by software-driven automation, real-time knowledge can provide the tools to mine available network margin and repurpose it to the benefit of Telia Carrier’s network operations as well as its end-user customers.

As optical networks become more dynamic and the evolution of coherent technology introduces a wider range of modulation schemes and baud rates, advances in software-driven automation, real-time performance awareness, and network adaptability create a practical framework for introducing a new level of optical layer efficiency and agility.

Figure 2: Telia Carrier use case illustrates the importance of residual margin in real-time performance awareness and the ability to maximize available link budget for network monetization.

Data Science and Policy-Driven Service Innovation

Discovery of the real-time state of the network to make intelligent and autonomous decisions is a practical automation use case proven to deliver tangible value for network operators. The ability to take predictive and/or prescriptive actions based on real-time knowledge extracted from collected network data provides the framework for dramatically reducing the number of manual tasks required in the management of multilayer, multidomain and multivendor networks. The building blocks for this type of automated, self-aware, adaptive network include:

  • The streaming of time-stamped data, called key performance indicators – this data detects any anomalies over time and essentially uses machine-learning algorithms.
  • Continuous learning via machine-learning-like components that entails understanding what happened and the reason why, learning to infer, and reporting the results.
  • An AI piece that enables the use of the inference and comes up with an automatic or policy-driven action to close the loop.

In the United States, CenturyLink has been active in this area, with recent thought leadership demonstrations of data science and policy-driven applications in an open network architecture. CenturyLink uses a combination of open protocols, software-defined technologies, and advanced networking automation
and intelligence.

Figure 3: CenturyLink PoC implementation of data science and automated intelligent network application for the MEF LSO architecture

The application of data science in the domain of telecommunications networks has the potential to handle real-world functions, including root-cause analysis, quality monitoring (that is, service level agreements), customer impact analysis, capacity analysis and planning, network forecasting, automatic problem detection, and predictive failure analysis and recognition. Open networking concepts and open application programming interfaces (APIs) enable automation on the extraction from source and insertion into the data lake.

In addition, APIs are essential in the configuration, activation and generation of reports on both physical and virtual devices that are the source of the data. In this data science use case, network functions virtualization (NFV) provides carriers and customers with the ability to place performance management functionality (for example, probes, source/sink clients) that can be used to analyze all layers of a network that supports any service. Visualization of the information can greatly enhance data science and provide insights that otherwise might not be realized.

CenturyLink’s work underscores the critical role that automation and machine learning can play in making intelligent decisions that drive value through significant advances in network optimization and efficiency. A predictive optimization algorithm has advantages over a reactive algorithm, because it can reduce the changes needed in the network and take network events into account before they show any signs in real-time measured data. Using a machine learning-produced link congestion prediction index, for example, multidomain orchestrators can perform more efficient network optimization compared with using traditional, measured, real-time network data.

Machine-learning algorithms also can be utilized by a multidomain context-optimized route engine (CORE) path computation engine (PCE), with network optimization facilitated by programmable policy engines, deep analytics, and automated pattern recognition and correlation.

Figure 4: Autotunable DWDM in action – Optics automatically tune to the port they are physically connected to, in this case the blue channel.

Autotuning Access Networks for 5G and DAA Network Evolution

Network pressure points change over time depending on market drivers, reinforcing the value of applying application-optimized automation
tools and techniques to address critical issues and drive immediate business value. Much more than putting a
finger in the dike, these automation tools can lay the initial foundation
for wider-scale deployments that accelerate transformation.

This has been the case for a leading network operator in Asia, which experienced the value of autotunable DWDM in the access network segment. The access network faces major shifts globally as new fiber deep architectures such as 5G mobile networks and distributed access architecture (DAA) in cable networks push fiber and DWDM systems deeper into the access plant.

These market shifts bring a range of new challenges for network operations:

  • Dealing with the proliferation of optical access points – where do I find enough field engineers with the right skill sets to deploy DWDM to all these new locations?
  • Coping with constrained space and power – DWDM now often needs to be deployed in locations that never were intended for complex transmission equipment.
  • Costs – Capex and opex costs are always important, but any savings in access networks are hugely amplified because of the large number of locations that can benefit from lower costs.

The optical networking industry is now bringing autotunability to DWDM optics, essentially enabling the 10G DWDM tunable optics used in access networks to learn their required wavelengths from the network. This enables field technicians to treat these optics as if they were simple grey optics. Because of the host-agnostic nature of these optics, the host systems also treat them as if they were grey. This capability enables the third-party host system to become DWDM enabled without needing any upgrades, which in turn removes the need for DWDM transponders and significantly lowers the capex of the network.

From an opex perspective, the benefits of this approach are quite obvious: Field technicians installing DWDM optics to support fiber deep-access networks don’t need to worry about dealing with a large number of fixed optics or tuning tunable optics to the specific required wavelength at a particular site. A single common autotunable optic is deployed at every location without any configuration by the installation engineer. In addition to simplifying the installation process, autotunable optics can help reduce sparing costs and the risk of accidental installation errors that can lead to expensive truck rolls.

The importance of automation in telco network transformation cannot be underestimated. It remains critical to managing the operational costs and network complexity that increasing data volumes and unpredictable traffic dynamics place on heterogeneous transport infrastructure – from fiber deep-access segments to the terabit-scale long-haul optical core. Though significant investments are being made in holistic, large-scale projects employing end-to-end SDN and NFV, many service providers are concurrently finding ways to tackle discrete operational challenges with these and other software-defined automation tools and machine-learning-like capabilities. This practical, easy-to-implement, application-based approach to network automation offers service providers a fast track to tangible business value, including accelerated time to market, simplified operations, reduced network complexity, and lower capex and opex costs.