Network operators are anticipating and in some cases enjoying the benefits of applying IT virtualization technology to customer premises equipment and network functions. But virtualization has its costs. For instance, it introduces additional complexity, creating the need for an orchestration layer with more sophisticated assurance capabilities, including data center and network analysis, encompassing physical and virtual resources in real time.
Advanced analytics are well suited to help operators meet this challenge. In particular, operations analytics (OA), is a good fit. It can help operators assure virtualized networks and services by driving automated, real-time systems that learn from data and autonomously adapt to new information. By intuiting connections and relationships, OA can proactively detect anomalies and then prescribe precise solutions.
Why Operations Analytics?
As we noted in a previous blog, analytics is the use of mathematics to understand data. What complicates this picture – and one of the reasons we have spent some time simply defining terms – is that advanced analytics comes in several flavors. Machine learning (ML), for instance, is a relatively mature branch of analytics, in which machines improve their ability to recognize patterns as they continue to be trained with additional examples – without having to be programmed to handle each new example or pattern.
By contrast, machine intelligence or artificial intelligence (MI/AI) is a somewhat newer field which improves upon ML by adding the ability to reason. Artificially-intelligent systems are capable of forming hypotheses from raw, disparate data to develop new, valid information that is not a direct result of data in the original data set.
Another category, operations analytics (OA) applies these techniques to business operations. More specifically, OA relates to historical and real-time business processes, including resource planning, service monitoring and service diagnostics. The broad OA functions relevant to network virtualization include the following (see also Figure 1):
- Collection at scale: collect event data in real-time and at massive scale from a variety of sources leveraging a big data engine;
- Real-time enrichment: enrich and fuse cross-functional data in real-time with other events and reference data – combining data in motion with data at rest;
- Analyze and predict with ML/AI: monitor millions of event time-series, apply ML for baselining, anomaly detection and real-time prediction with models built via AI algorithms for actionable intelligence; and
Drive action or decision: prescribe actions and integrate with downstream systems to perform a network or operations function.
Virtualization and OA
Network functions virtualization (NFV) debuted about five years ago when many service providers were engaged in cloud initiatives. It aims at a number of goals, including cost reduction, speed, agility, innovation and improved services. Software-defined networking (SDN), especially through its separation of control and data planes, complements and enhanced NFV-enabled infrastructure.
From the start, NFV was envisioned as interacting with operations support and network management systems (OSS/NMS). The combination of service provisioning and management along with new elements, such as network controllers and cloud managers, underscores the need for that interaction. Within an SDN/NFV framework, it is the orchestration layer that handles the complex tasks of bridging physical and virtual resources.
For SDN/NFV to reach its potential, however, it needs automation. Programmable networks generate considerable amounts of data, which create the potential for more intelligent, closed-loop decision-making. Able to adapt to circumstances and draw new connections and insights through real-time data processing, OA becomes integral to orchestration and advanced OSS.
More Precise and Timely
The link between OA and programmable or virtualized networks in practice occurs in hybrid scenarios. Implementations draw upon legacy and new network elements for OA-driven anomaly detection and prescriptive decisions. It is the real-time capabilities of OA that offer a decisive advantage. Whereas the status quo approach may generate static metrics, applied universally and misaligned with circumstances; OA leads to more accurate and timely data-driven actions.
In a subsequent blog, we will address several use cases that demonstrate the applicability of OA to resource allocations within virtualized networks, in both the cloud and access network.