Forecasting Capability in Cisco Catalyst SD-WAN


Organizations are more and more adopting software-defined vast space networks (SD-WAN) to boost community efficiency, scale back prices, and enhance total connectivity.

Utilizing synthetic intelligence (AI) and machine studying (ML) for IT operations (AIOps), Cisco SD-WAN enhances and simplifies community administration through the use of predictive analytics based mostly on AI and ML methods. The result’s a proactive instrument to deal with potential community points earlier than they degrade community and utility efficiency.

Options desired by networks operators for such proactive actions embrace:

  • Predictive Path Suggestions (PPR), which suggests most popular paths for numerous utility teams at every web site inside an overlay based mostly on long-term modeling of path high quality.
  • Bandwidth forecast for capability planning, giving operators insights into doable future community utilization based mostly on intensive previous utilization patterns.
  • Anomaly detection for community KPIs (tunnel loss, latency, jitter), utility utilization patterns with particular person websites, and consumer utility utilization profile.
  • Software modeling to assist community operators higher perceive the affect of rolling out new purposes within the overlay to allow them to implement the proper insurance policies for greatest efficiency and minimal affect.

In my final weblog, we mentioned PPR and demonstrated the way it offers operators the most effective efficiency for purposes on their cloth. In as we speak’s submit we’ll delve into Bandwidth Forecast. To completely leverage the advantages of SD-WAN, efficient capability planning is essential to assist guarantee optimum community efficiency, much less downtime, improved value management, extra seamless operations, and a superior consumer expertise.

The Bandwidth Forecast function takes a complete strategy to offer correct predictions of circuit utilization, offering visibility into which circuits are more likely to breach the capability threshold based mostly on the anticipated utilization. This helps community operators monitor utilization traits on the circuits and supplies capability planning for the overlay.

The forecasting is based on the RX/TX bandwidth info of circuits within the WAN cloth. To make sure insights use underlying long-term traits, the circuit utilization information is aggregated as day by day information factors whereas monitoring day by day Min/Max ranges. Aggregated information over prolonged intervals is used to generate a forecast for as much as three months sooner or later.

Varied different options inside this information set might be additional leveraged to boost forecast accuracy. These embrace:

  • Kind of circuit (e.g., MPLS, personal web, LTE)
  • Kind of purposes utilizing the circuit (i.e., high 10 purposes and their respective quantity)
  • Variety of customers on the web site served by the circuit
  • Regional vacation listing and bandwidth info options

To realize the most effective forecast doable, a mixture of frequent predictors and people based mostly on deep studying methods are used to generate extra dependable and strong forecasts.

Pre-processing of interface statistics for coaching and inference pipeline (Click on picture to enlarge)

Forecast high quality is constantly monitored for accuracy. If any information or mannequin drift or deviation from anticipated outcomes is noticed, retraining of the mannequin is triggered based mostly on up to date information units to enhance mannequin accuracy. Moreover, forecasts are assessed for long-term overestimation or underestimation, making certain that it faithfully predicts the bandwidth to help community operators in capability planning and decision-making course of.

The Bandwidth Forecast function in Cisco SD-WAN Analytics helps give community operators a greater understanding of the next:

  • Development Traits: By analyzing historic information offered facet by facet with the forecast, organizations can establish patterns and anticipate future bandwidth calls for. This empowers them to plan for anticipated progress with out disruptions.
  • Seasonality: Lengthy-term visibility into seasonality of utilization over the historic interval over which the coaching information set is derived from. The day by day, weekly, and month-to-month seasonality can be factored in whereas making the forecast and the sample continues into the forecasted information factors.
  • Surge: Though visibility is offered into historic surge utilization within the overlay so community operators can correlate it to world occasions (e.g., Black Friday) or inside occasions (e.g., firm all-hands video stream), the mannequin is efficient in minimizing the affect of such information factors whereas making long-term forecasts.
  • Min/Max Band: The day by day information factors for forecast has three elements, Min, Imply, and Max. The forecast is offered with emphasis on the day by day imply worth whereas nonetheless exhibiting a Min/Max Band in order that the community operators can get insights into utilization spikes inside the day.
  • Mannequin/Forecast Efficiency: Historic utilization information is offered together with the previous forecast information factors for a fast visible comparability of how the forecast carried out in opposition to precise recorded values previously.

Consumer interface

The Bandwidth Forecast function might be activated for a particular overlay within the Catalyst SD-WAN Analytics Dashboard. This seems below the “Predictive Community” tab. Customers can select the circuits within the overlay for the forecast technology.

A desk of circuits with all associated metrics similar to web site or supplier data, RX/TX bandwidth, and complete utilization is displayed, serving to customers choose the circuits for which they wish to visualize Bandwidth Forecast particulars. The minimal information set requirement for forecasts to be generated is 12 weeks of historic day by day information factors for every circuit.

The workflow is topic to the next:

  • The desk exhibits solely circuits configured on bodily interfaces and this can exclude any circuits configured on logical interfaces (e.g., sub-interfaces, loopback, dialer-group).
  • Default sorting relies on descending order of RX/TX bandwidth, which helps bubble most closely used circuits to the highest of the desk. The chart show is used to point out the forecast for the Prime Circuit.
  • Customers can choose every other circuit by clicking on the checkbox.
  • Customers can search and type as they want to isolate particular circuits of curiosity.
Desk of circuits and their metrics [Click image to enlarge]
Bandwidth Forecast for chosen circuit exhibiting precise and predicted (dotted) values [Click image to enlarge]


Correct bandwidth forecasting is important in capability planning. One key metric is the accuracy of the forecasted bandwidth necessities. A profitable forecast ought to carefully align with the precise capability targets for your small business. The present answer computes imply absolute share error (MAPE) and imply absolute scaled error (MASE) scores along with monitoring percentiles. Any of those can be utilized because the optimization goal for the predictors used. The selection of goal metrics for the predictors might be specified as per the wants for a particular overlay or use case.

By precisely predicting bandwidth necessities, organizations can optimize site visitors routing, provision acceptable hyperlink capacities, handle QoS successfully, plan for scalability, and guarantee adherence to SLAs. This proactive strategy allows companies to leverage the total potential of SD-WAN, delivering enhanced community efficiency, improved consumer experiences, and the power to adapt to altering enterprise wants. As organizations embrace the digital transformation journey, incorporating bandwidth forecast in SD-WAN capability planning turns into a key technique for fulfillment.


Be taught extra about Cisco SD-WAN, Analytics, and WAN Insights:

Cisco SD-WAN Analytics

Cisco SD-WAN vAnalytics Animated Explainer Video