Partner POV | NETSCOUT Omnis AI Insight: Unleashing AIOps

This article was written and contributed by, NETSCOUT.

Executive Summary

Communication Service Providers (CSPs) have the potential to pair high-quality data from the network with AI inferencing engines to transform all aspects of network, security, and IT operations: hence, AIOps. This early-stage market promises to yield business outcomes that were once thought impossible just a few years ago.

Recent market developments have been driven by a variety of factors including faster computational GPUs, a tsunami of data, and large multi-dimensional data models. We are now entering a market phase that promises to deliver billions of dollars in revenue and cost savings. Applied correctly, AIOps will deliver higher-quality outcomes and transform the way planning, operations, and customer support teams conduct their work.

But given the heightened demand and buildout of 5G SA (Standalone) and edge technologies, there is a notable shift toward needing to access data in a multi-vendor, cloud-native network, particularly as the radio access network becomes more disaggregated. As a result, many CSPs struggle with the data paradox – the very availability of vast volumes of granular data in their networks making it ever more challenging to extract high value insights. And that is even before considering how quickly insights can be turned into appropriate corrective action.

NETSCOUT, with its strong background in IP network monitoring and analysis, is positioning itself to address these emerging opportunities and challenges. The company is introducing new products designed to simplify the management of disaggregated network functions across multi-vendor environments. These can be deployed as part of NETSCOUT's own offerings or integrated into workflow pipelines for use with third-party solutions.

The new NETSCOUT Omnis™ CSP AI Insight platform integrates an Omnis CSP AI Sensor and AI Omnis CSP Streamer aimed at supporting both AIOps and SecOps teams. The solution extracts data at or near the source and generates enriched metadata for advanced use cases. These include: gaining insights into customer behavior and preferences, monitoring spectrum depletion through heavy users in fixed wireless access (FWA), providing assurance for 5G SA service slicing, enhancing security with improved threat detection, boosting net promoter scores (NPS), and enabling various monetization use cases, such as traffic planning, security, and prediction. The platform also supports enhanced service offerings through network exposure APIs and helps identify fresh marketing opportunities, among other applications.

As networks become more intricate with 5G SA and cloud-native technologies, the ability to harness high-quality operational data, and apply AI-driven insights will be increasingly critical for CSPs aiming to maintain their competitive edge, operational efficiency, and superior service delivery.

Introduction

The emergence of 5G Stand Alone and cloud-native technologies has transformed network service delivery, providing greater flexibility and scalability. However, this shift has also introduced increased complexity, driven by disaggregated network functions, enhanced encryption layers, new packet message formats, and growing volumes of subscriber data traffic. As a result, communication service providers (CSPs) are seeking higher-quality operational data to better optimize network and service availability, performance, and profitability.

At the same time, however, operators are keen to see optimization, performance and quality of service not as network issues but through the prism of customer satisfaction.

Appledore Research has conducted hundreds of briefings over the past year with CSPs and key suppliers to the AIOps market. Our findings reveal that the major thrust for most CSPs can be summarized in five critical areas of their business (figure 1). Forty-five percent (45%) of respondents in a sampling of eighty-five global operators find that network observability and planning will account for AIOps investments over the next 2 years. The predominant family of use cases is noted in the graphic below.

Within Operations, timely and accurate data access enables service providers to proactively identify network and security issues before they escalate to trouble tickets. In this context, the mean time to identify issues becomes a key metric and is potentially more significant than the mean time to resolve, especially when AI is applied alongside relevant datasets. This proactive approach not only minimizes disruptions to business operations but also ensures a seamless experience for end-users.

The critical success factor for AI initiatives is transforming raw information into strategic assets that drive significant business value. The true power of AI emerges when it can sift through enormous datasets, identifying patterns and extracting knowledge that human analysis might overlook.

Becoming Data-Driven

The Data Paradox

Data plays a pivotal role throughout the AI lifecycle: it trains ML algorithms, validates their efficacy, and evaluates AI-generated outcomes. During the validation phase, data is crucial for refining and optimizing models. It is imperative that data be meticulously classified and labeled following its identification and gathering process. The foundation of any successful AI endeavor lies in the quality, accessibility, and management of its data resources. From training models to validating outcomes, data integrity and availability are indispensable in achieving meaningful AI-driven insights and applications.

But a significant challenge frequently encountered by implementers of AI-driven projects in telecom is the data paradox: although petabytes of data are being generated every day, CSPs find it difficult to gain useful insights on subscriber experience. The data value paradox highlights the challenge of transforming raw data into actionable insights that can drive strategic decision-making and operational improvements. The data demand paradox is a situation where CSPs are collecting data at a parabolic rate, often outpacing their ability to analyze and utilize it effectively. This creates a constant tension between data acquisition and data utilization capabilities.

Practical challenges in network data

The 5G SA network generates rich contextual data, but it must be properly identified and classified to be usable in the AIOps pipeline. To prevent suboptimal outcomes and wasted resources, CSPs must rethink how they apply best practices in data harvesting and its integration into the AIOps process.

Several challenges arise in mining network data:

• Data collection often lacks real-time capability to support modern, software-defined services.
• The vast amounts of network-level data require considerable time to assemble and analyze.
• Data is not always provided in a standardized, open format that is easy to consume.
• Application quality of experience, by user and by service, often receives insufficient attention.
• There is limited correlation between user identity and the data, services, or applications they use.
• Understanding how network services are utilized at the user or enterprise level is limited.
• Security issues affecting both the network and user experience remain opaque.
• An end-to-end view of the network is often lacking, providing only partial visibility.
• Business intelligence on device and application behavior within the network is insufficient.

Accessing data in a 5G SA environment presents additional barriers, such as dealing with TLS 1.3 encryption on the Service-Based Interface, complex cloud-native network architectures, and the vast amounts of user plane data that need to be processed. These obstacles must be overcome to achieve effective data acquisition.

Beyond acquiring the data, it requires a refinement process similar to refining gold. The data must be contextualized to show the network quality provided to each subscriber — understanding the "who, what, when, and where" along with the network conditions affecting service quality. A complete end-to-end view, including RAN, MEC, and CORE visibility, is essential, with this data packaged and ready for consumption by the AIOps pipeline.

The long Game: unleashing AI to drive Automation

AIOps plays a crucial role as a facilitator of network automation, but telecom is in a transitional phase between human-assisted and partial automation, according to our observations of ongoing deployments.

Figure 2 provides a snapshot of the industry's progress towards automation, emphasizing that human oversight remains integral at this stage:

Fully autonomous operations, where networks operate "hands-free," are considered a distant prospect - requiring more robust prediction models to instill confidence in autonomous network activities.

Network Automation Software (NAS) facilitates the implementation of control loops, recognized as highly effective methods of automation that are widely adopted across various industries. Central to their operation is defining outcomes as intent, with algorithms adjusting implementation details to achieve these objectives. Figure 3 illustrates the process of data ingestion, serving as a foundation for data observability platforms:

Following ingestion, machine learning algorithms optimize anomaly detection within multidimensional datasets. The prediction engine then leverages this analysis to forecast future outcomes within relevant contexts which then suggests automated remediation actions or escalating issues to human operators for manual intervention.

AIOps evaluates the attainment of intent, identifies potential future issues, distinguishes between normal and abnormal states, and pinpoints root causes for orchestration. Control loops coupled with cloud-native infrastructure, intent-driven operations, and AIOps, are pivotal for the efficient operation of public clouds and are increasingly indispensable in telecom networks, albeit still in initial stages. Nonetheless, the industry is optimistic about advancements, with suppliers focusing on intent-based and algorithm-driven orchestration, active assurance, machine learning, and multilayer dependency graphs. These innovations lay the groundwork for more sophisticated automation capabilities.

AIOps also provides a new pathway to increased revenue streams and cost reduction to the CSPs. It allows for increased operational efficiency though automation while providing new go-to-market ecosystem – user centric and data-driven open platforms to onboard ecosystems services via Open API's.

The edge market offers potentially new opportunities for CSPs to generate new sources of revenue, leveraging AIOps. As other industries such as retail, hospitality, and media look to move workloads to the edge of the network, AI inferencing can be applied to generate new business models and new sources of revenue for CSPs as enhanced services to these same industries.

NETSCOUT Innovation: Smart Visibility – AI Applied to Packet Level Data 

NETSCOUT processes high-value data in IP networks, providing comprehensive insights across multiple areas, including networking, applications, and security. With extensive deployments in mobile networks, NETSCOUT analyzes both the control plane and user plane, extracting packet data across various physical and virtual environments. The company's focus is on delivering scalable Layer 7 detection and addressing the challenge of scaling packet data analysis.

In the network security domain, NETSCOUT has been increasing its market share. Traditional tools often face difficulties in detecting modern attacks, particularly those targeting applications. NETSCOUT's security products use packet data to address complex, multifaceted threats. The integration of AI sensors and data streamers further enhances detection capabilities, highlighting the importance of their data-driven approach.

Figure 4 depicts NETSCOUT's Omnis CSP AI Insight architecture, which can ingest hundreds of gigabytes of data in real-time. The data undergoes ingestion processes such as aggregation, filtering, and correlation, followed by initial telecom domain-based AI analytics. This results in a curated data stream that drives an outboard AIOps engine. The insights from this data stream, combined with other data sources, fuel AIOps.

The architecture includes open API interfaces that enable the AIOps engine to specify the required data and the level of curation for each use case. This setup allows feedback from the AIOps engine to the AI Streamer and AI Sensors, enhancing operational awareness within the AIOps data pipeline. Additionally, this data empowers communication service providers (CSPs) to create targeted and effective marketing campaigns, opening new opportunities for revenue generation.

Integrating AIOps solutions into the Telecom sector can be challenging, and NETSCOUT helps customers mitigate the risk of subpar business outcomes by delivering network data "tuned" for the AIOps pipeline This is achieved through comprehensive visibility of layer 3–7 data plane capture for every user on the network, with full correlation of control plane traffic all packaged together and curated for the customer's AIOps engine.

NETSCOUT's position is distinctive. Instead of relying solely on its own AI engine, NETSCOUT recognizes the rapidly evolving AI/ML landscape with continuous innovations from specialized AIOps vendors. This allows NETSCOUT to focus on their core strength: delivering the highest quality network data in the most efficient form to advanced AIOps engines. This strategy aligns with the TM Forum's guidelines on Autonomous Network Architecture.

NETSCOUT has also taken extra steps to ensure the data provided via the AI Streamer can comply with international standards such as The European Union GDPR which protects personal data. This includes any information that can directly or indirectly identify an individual, such as IP addresses, and even pseudonymized data.

This extension to the NETSCOUT portfolio should enable them to extract additional value from the network data they collect, benefiting both their existing customer base and new clients. It also opens new adjacent pathways within the CSP community to leverage this data for monetization purposes. Regarding monetization, the concept of programmable data on the fly aligns closely with the direction the Network Exposure via Open API (CAMARA) group is pursuing. The expanding partnership ecosystem, which allows partners to leverage Omnis CSP's AI Smart Data, will act as a catalyst to broaden its market presence as a valuable source of network metadata.

High Impact Use Cases

NETSCOUT's high-fidelity data, curated through the AI Streamer, unlocks a wide range of use cases for the AIOps engine, including subscriber experience optimization, network analytics and automation, threat detection, and monetization/personalization opportunities.

Figure 5 illustrates several examples of these use cases, which span across various industries. For instance, cable providers transitioning to virtual environments can utilize these insights to assess system readiness, using key metrics to inform significant investments in nationwide virtual rollouts and new architectures.

Example: Automate VIP Traffic

The AI Sensor can be configured to efficiently analyze user plane data at scale, distinguishing between VIP and non-VIP traffic. This capability is essential in developing markets like India, where there are tens of millions of subscribers, and not all data holds the same level of importance. For example, while music streaming may be considered lower priority, a special telecast with interactive voting would be classified as VIP traffic.

This VIP classification helps the network differentiate between standard and critical traffic, allowing for focused monitoring of important activities without overwhelming the system with non-essential data. The system segments data into two categories: non-VIP, which covers broad subscriber activities and volumes, and VIP, which collects detailed key performance indicators (KPIs), session data, and packet recordings. This method enables efficient data management and ensures network resources are allocated to more critical monitoring.

Use cases for this technology vary based on the needs of Tier 1 operators, such as managing contract overages in fixed wireless access points or improving customer experience. The system is flexible, supporting hybrid architectures where network data can be integrated with insights from external sources, like social media, for a comprehensive view of user behavior.

The system architecture balances traffic across sensors based on subscriber identity, ensuring detailed analysis of VIP users while maintaining broader monitoring for non-VIP traffic. This results in two parallel streams: one for non-VIP traffic focused on volume, and a more detailed stream for VIP traffic, capturing all relevant data. The system automates this selective data collection, improving efficiency and reducing unnecessary noise in the analysis.

The system provides high-quality metadata for advanced analytics and automation, compatible with both legacy and emerging automation tools. It supports various network types and generations and offers detailed insights using open APIs and standard formats.

In one live deployment, the system reduced data volume significantly. Initially, forty-one terabytes of user plane data were collected, but with VIP classification, only 166 gigabytes were flagged as VIP, resulting in major data savings. This allows for detailed analysis of VIP user behavior, such as application usage and network mobility, without overburdening the system with lower-priority data.

For VIP users, everything from application usage to network handovers is carefully tracked, offering a thorough view of their network experience. This level of detail extends to automatic user categorization, enabling the system to dynamically adapt to network changes. For example, different analysis tiers can be applied on specific days, ensuring effective monitoring of both VIP and non-VIP segments without constant high-volume data processing.

The system's flexibility allows operators to implement custom filters or code for unique deployment scenarios, offering a tailored solution that meets the specific needs of different markets and operational conditions.

Conclusion

The goals for AIOps are clear and compelling: dramatic improvements in automation, better experiences for customers. The obstacles to achieving those goals are substantial – but not insurmountable. Capturing vastly increasing amounts of network telemetry in real time, transporting that (or processing it close to source), extracting key insights and relating that to customer experience all depend on a range of sophisticated innovations and domain knowledge.

NETSCOUT's strategy in AIOps is to provide highly programable software that delivers efficient, use case driven sets of "Smart Data" to an AIOps engine. It achieves this through the AI Sensor (network data acquisition engine) and AI Streamer (data processing, curation, and metrics engine). Its approach reduces overall data bloat, storage costs and expensive GPU processing time for insignificant data, thereby minimizing the time required to detect emerging network issues.

Processing multi-vendor network data is a key consideration for CSPs in AIOps – another NETSCOUT strong point.

The adoption of AIOps opens new opportunities for market expansion, and collaboration with additional business data platforms to enhance service offers. The trend towards cloud migration also presents opportunities to provide cloud based AIOps applications, positioning NETSCOUT as a leader in providing the highest quality of curated network data to feed those cloud based AIOps engines.