WAN Insights

A predictive network is a network that leverages a predictive analytics process using historical data to forecast future behavior to determine the probability of an issue occurring or reoccurring on that network. Then, the forecast is capable of generating recommendations that, if applied manually or through automation, reduce the probability of the issue ever happening and, therefore, reduce the impact on the user's digital experience.

Predictive analytics is the application of algorithms applied to a model in an effort to identify the potential of future outcomes based on historical data.

Creating a predictive analytics model involves running different algorithms to identify recurring patterns in sizeable transactional data sets.

While machine learning is not technically a form of predictive analytics, it can be leveraged to enhance a predictive analytics model's ability to detect patterns, forecast outcomes, and recommend actions over time.

Predictive analytics is forward-looking data analysis that identifies the likelihood of future outcomes based on historical data. Diagnostic analytics is reactive and designed to determine changes that occur in a specified period and is typically used retrospectively to understand why an event or trend occurred.

Enterprises are increasingly adopting cloud-based services, such as SaaS applications, which in turn increases reliance on the Internet to deliver WAN traffic. The use of traditional multi-protocol label switching (MPLS) services makes sub-optimal use of costly backhaul WAN bandwidth, given that many critical applications and services are no longer internal in data centers. As a result, enterprises are migrating to hybrid WAN architectures and SD-WAN technology that combines direct Internet access (DIA), IP VPN tunnels, and traditional MPLS circuits.

Over the last several years, a robust market of SD-WAN vendors and managed SD-WAN providers has arisen to meet this need. SD-WAN has been architected to provide a certain amount of resilience, leveraging automated actions. A number of these automated functions are able to track network and path characteristics of the data plane tunnels between SD-WAN devices and use the collected information to compute optimal paths for data traffic, automatically redirecting data traffic to the best available path in the event of an issue occurring. This function, while automatic, remains reactive, only able to initiate the change when conditions deteriorate to the point at which the user's performance is impacted.  

Predictive analytics builds on this reactive functionality by taking those characteristics, including packet loss, latency, and jitter, and applying a broad range of statistical time-series methods to predict probabilities of traffic disruption for different applications and use these forecasts to provide recommended path selections to the network, in order to avoid the probable issue occurring and impacting user performance.

The goal of the ITOps team for any existing and new site within an SD-WAN network is to provide the best application performance while keeping the circuit cost low. However, the volume of data points across thousands of sites, networks, paths, applications, and users is impossible to assess manually and analyze. As a result, most organizations have a reactive approach to delivering user experience, only focusing on issues as they occur, and are unable to move away from a firefighting mode of operations.

A reduction in the number of incidents and, subsequently, the number of times ITOps has to react to an issue allows them to move beyond a complaint-driven, fire-fighting engagement to a data-driven decision-making process. This facilitates a proactive engagement model that allows network conditions that may otherwise have gone unnoticed to be addressed before they reach a noticeable user-impacting level.

This becomes applicable throughout multiple levels of ITOps, enabling cross-skilling throughout the organization, providing less experienced engineers with enough information to make informed decisions, and ultimately freeing up time for more experienced engineers that can be utilized for strategic actions.

By continuously observing the behavior of selected applications across every network path over a rolling period, a predictive analytic system is able to analyze historical performance across all SD-WAN network sites and identify quality issues for each application category (for example, Office 365 or Voice).

When the system identifies that a quality issue could be avoided by a change to the current network configuration, it generates a recommendation, typically an alternate path to route that site’s outgoing traffic, for a given application category.