Enabling Automation in Telecom Networks

By Kamakshi Sridhar, Ph.D. VP Automation Innovation, Sandvine

Kamakshi Sridhar, Ph.D. VP Automation Innovation, Sandvine

Traffic growth driven by streaming video, social networking and newer bandwidth hungry gaming applications has led to increasing levels of network congestion along with higher expectations of user Quality of Experience (QoE). Users are no longer willing to accept lower video resolution or even occasional choppy over-the-top (OTT) voice quality. Operators are upgrading their network infrastructure to 5G, FTTH, DOCSIS3.1, etc. to support high throughput, low latency, low jitter, and low loss required for various user applications. Networks are getting increasingly more complex to manage due to a mix of legacy and new technologies that must coexist and work cooperatively to serve the end user. There are a large number of parameters that need to be fine-tuned over a wide range of time scales for stable system operation. Operation teams need to address issues such as, which parameters should be fine-tuned? How often should they be fine-tuned? Will incorrect parameter fine-tuning result in unintended network instabilities? Operators are looking for innovative ways to monetize their infrastructure by exploring newer business models. NFV, SDN and Cloud Services delivery allow operators to lower their operating costs, defer their CAPEX, and deliver innovative services in an agile manner. Automation is a key enabler that helps unlock the full potential of these technologies to help operators deliver the highest user QoE at the lowest cost per bit.

Automation is a set of processes that helps operators accomplish business objectives such as lowering the network operating cost by squeezing out operational inefficiencies, deferring capital expenditure by sweating out existing network assets, lowering the cost of addressing user application QoE issues through proactive network fault and application diagnostics and providing insights that predict user churn. Automation processes allow the network to adapt its operation, often in near-real-time, in response to traffic variations with little or no manual intervention. Automation exists in parts of the network, such as power control in wireless, fast reroute in MPLS networks, etc. These work well. What is the new is that operators are now extending the scope of automation to include the user, content sources, and multiple protocols across multiple networking domains. Uncertain of how a network would operate with automation, and yet keenly aware of the promise and benefits of automation, operators are cautiously building and incorporating automation processes in their networks.

“Automation is expected to play an increasingly larger role in telecom operator networks. Well-designed processes with systematic execution can mitigate the complexities of automation and ease its adoption within operator networks.” 

Enabling automation involves integrating new decision making algorithms and processes in the networking infrastructure, operational workflows, and customer care workflows. These processes must coexist and engage with the existing network protocols while adding additional capabilities that manual processes may not match in scope or responsiveness. The design of automation processes is greatly simplified if the following issues are addressed early on: what does the automation process aim to accomplish? Which organization within the operator network is it intended to help? Who will use the outcomes of the automation effort and how will they be used? Establishing an end-to-end workflow is critical to the success of automation.

An automation workflow defines the various steps that ingest and filter the raw network data, process it, and act on it so that the outcomes can be consumed and evaluated for benefits. The business objectives driving automation establish the success criteria, and hence the outcomes of the end-to-end workflow. The business objectives may translate to an ‘Intent’ for anautomation engine that is expressed in network-specific requirements. For example, a desired ‘Intent’ may be that latency must be less than 40 milliseconds for a gaming application. The automation engine comprises of algorithms, inline or offline, that modify the network parameters appropriately to achieve the desired ‘Intent’. The outcomes of the automation engine may constitute either the entire set, or a subset, of the end-to-end workflow outcomes. Initially, automation may require manual assists to ensure that the data pipeline works as expected and that the outcomes are useful to the operator.

Algorithms are the heart of the automation engine. They are built for a specific network domain and designed for a specific automation application. Algorithms may be comprised of physics-based models, empirical-based models, machine learning models or variants thereof. Keeping models simple early on ensures that the models can be verified early on when invoked by the end-to-end workflow. The models should account for the traffic variation dynamics and subscriber variation dynamics so that the automation process is responsive to allow for timely actions. Safeguards must be built in to protect the network from unexpected outcomes of the automation process, particularly when the automation mechanism is in line in the data path and hence impacts the user QoE directly. Iterative testing with each model upgrade saves debug time and builds confidence in the outcomes. New features can then be added gradually to build more complex models that produce more detailed outcomes. Packaging the components of the workflow so they can be installed, deployed, configured, and launched easily greatly speeds up the deployment of the automation process.

Good quality traffic data, fresh in its periodicity, with rich contextual information, is critical to the automation process. Better data leads the algorithms to produce more accurate insights. It leads to higher confidence in applying specific network actions which results in trust-worthy automation outcomes. The automation pipeline that comprises of compute and storage units must be architected and dimensioned to be scalable across the network where it will be deployed. Good visualization tools, integrated into the network operations workflows, can be valuable in interpreting the outcomes and providing unique perspectives to different network operator personas. Minimal manual intervention allows the automation engine to be exercised frequently to produce repeatable, measurable, and ultimately reliable outcomes.

This brings us back to the automation objectives. Automation aims to achieve a certain business outcome. ROI models help assess the impact of automation in realizing those outcomes. If the automation process has a measurable positive impact on each of its value drivers, it will lead to a positive return on investment (ROI). Exercising the automation workflow early on helps quantify the automation benefits and evaluate the ROI early on. These early ROI estimates can serve as a benchmark for further improvements with automation. Statistical testing with a sufficiently large and diverse sample size will enable operators to draw unbiased inferences that validate intuition and lead to the acceptance of the automation outcomes. Where possible, verification of the outcomes through corroboration with other independently available metrics can further establish confidence in the outcomes of the automation process. Periodic re-evaluation of the automation outcomes preserves the integrity and accuracy of the automation process as the operator network evolves over time.

Ultimately, the key metric for evaluating the success of automation is acceptance by the operator. What enables automation to succeed is human expertise. Automation requires personnel skilled in scoping the opportunity and defining the problem, understanding the domain phenomena and developing appropriate algorithms, building the workflow components with high quality, and adequately testing and validating the outcomes. Automation is expected to play an increasingly larger role in telecom operator networks. Well-designed processes with systematic execution can mitigate the complexities of automation and ease its adoption within operator networks.

tag

ROI

Read Also

The 5G Container Ship: Loaded With Sliced Beams and Liferafts

The 5G Container Ship: Loaded With Sliced Beams and Liferafts

Mansoor Hanif, Director of Converged Networks and Innovation, BT
Is your Enterprise's Wireless Network Ready for 5G?

Is your Enterprise's Wireless Network Ready for 5G?

Todd Landry, Corporate Vice President of Product and Market Strategy, JMA Wireless
Applications of Text Analytics

Applications of Text Analytics

Rajkumar Bondugula, Ph.D., Principal Data Scientist / Sr. Director, Equifax