Excessive Alert Noise: Cause, effect, and solution

November 6th, 2019

With an exponential growth in the IT sector over the last few years, traditional operational tools and process  isn’t enough to stay ahead of the market. Problems/anomalies are treated as ‘events’. Each of these events triggers an alert in the system leading to separate incidents that require individual resolution. With an increase in data, hybridization, operational tools, countless metrics, there has been a corresponding increase in alert volume. This causes inundation of high volume and variety of log data, usually with multiple false and redundant alerts.

About 40% of IT organizations see over a million event alerts a day, with 11% receiving over 10 million alerts a day.

Most IT teams today operate in disparate silos, often unaware of the assets they have, their utilization or inter-dependence thereby compounding the problem. 

Why is there an excess of alert noise?

Some of the common reasons for an increasing volume of alert noise are:

  1. Lack of stack awareness 
  2. Static thresholds 
  3. Alert Storms

Lack of stack awareness

Traditional legacy systems process this differently using approaches that solely rely on signature/footprint matching. This does not allow for Machine Learning capabilities to perform impact analysis and correlation of alerts/events from multiple stack elements.

Static thresholds

Static thresholds are unable to take into account the dynamic nature of IT workloads. This creates alerts at pre-established levels, that no longer works for a majority of the workloads leading to an excessive number of alerts. Being unable to identify and create contextual awareness of where to disabled alerts and where to increase alert capacity proves to be a barrier.

Alert Storms

Outages both planned and unplanned stir up alert storms. Network disruption causes employees, remote users, and devices to disconnect leading to a high volume of unwanted alerts. 

Alert Noise is estimated to cost an average of $1.27Million per year to companies.

How does CloudFabrix help with alert reduction?

The CloudFabrix AIOps platform  uses combination of user configurations and advanced AI/ML algorithms such as correlation , anomaly, forecasting etc to reduce alert volume through grouping, suppression and prevention. 

AIOps has been implemented by 60%of organizations to reduce noise alerts and identify real-time root cause analysis.

Rule Based – > AI/ML and Analytics Based Approach

Instead of relying on manual tagging and rule based grouping, CloudFabrix uses time based and asset dependency based automated grouping of multiple alerts into actionable problems. It further uses predictive analytics thereby reducing alert noise by a significant number. 

Static Thresholds → Dynamic Thresholds

Static thresholds ignore dynamic nature of IT workloads and create alerts at per-established levels, which won’t work for the majority of the IT workloads that are dynamic in nature. This results in excessive number of alerts.

To address this problem

  1. Granular Controls: Provide granular alert controls to tune telemetry collection interval. And to minimize the alerts caused to metric fluctuations we provide hi-watermark, lo-watermark and minimum occurrence controls.
  2. Dynamic Thresholds: Dynamic thresholds establish a baseline for every metric and raise an alert only if the metric is deviating from baseline.

Identify heavily utilized assets where alerting should be disabled or more capacity should be added.

Alert Storms →  Actionable Incidents

Alert Storms can occur anytime, but more so during unplanned outages , planned outages and cascading alerts 

  1. Planned Outages: With our platform, alerts can be configured to be ignored during planned outages like patching, backup or maintenance. In addition to this, we are able to automatically exclude network device access ports from monitoring, as this can cause an excessive number of unwanted alerts, whenever employees, remote users, phones etc. connect/disconnect from the network.
  2. Unplanned Outages: and device fluctuations or flapping situations cause alert storms, which we detect automatically  and suppress the alerts during unplanned outages like network disruption or device unavailable events.
  3. Cascading Alerts: this happens when a device/component fails resulting in alerts from other parts due to interdependence or lost connectivity between the monitoring system and the dependent devices. These deluge of alerts are often pointing to the same underlying issue. These sort of alerts can be grouped together if the system has knowledge of the interdependencies and can identify the underlying root cause issue. 

Please feel free to reach out to us in case of any questions.