IBM Databand: Self-learning for anomaly detection

Almost a year ago, IBM encountered a data validation issue during one of our time-sensitive mergers and acquisitions data flows. We faced several challenges as we worked to resolve the issue, including troubleshooting, identifying the problem, fixing the data flow, making changes to downstream data pipelines and performing an ad hoc run of an automated workflow.

After the immediate issue was resolved, a retrospective analysis revealed that proper data validation and intelligent monitoring might have alleviated the pain and accelerated the time to resolution. Instead of developing a custom solution solely for the immediate concern, IBM sought a widely applicable data validation solution capable of handling not only this scenario but also potential overlooked issues.

That is when I discovered one of our recently acquired products, IBM® Databand® for data observability. Unlike traditional monitoring tools with rule-based monitoring or hundreds of custom-developed monitoring scripts, Databand offers self-learning monitoring. It observes past data behavior and identifies deviations that exceed certain thresholds. This machine learning capability enables users to monitor data with minimal rule configuration and anomaly detection, even if they have limited knowledge about the data or its behavioral patterns.

Databand considers the data flow’s historical behavior and flags suspicious activities while alerting the user. IBM integrated Databand into our data flow, which comprised over 100 pipelines. It provided easily observable status updates for all runs and pipelines and, more importantly, highlighted failures. This allowed us to concentrate on and accelerate the remediation of data flow incidents.

Databand for data observability uses self-learning to monitor the following:

• Schema changes: When a schema change is detected, Databand flags it on a dashboard and sends an alert. Anyone working with data has likely encountered scenarios where a data source undergoes schema changes, such as adding or removing columns. These changes impact workflows, which in turn affect downstream data pipeline processing, leading to a ripple effect. Databand can analyze schema history and promptly alert us to any anomalies, preventing potential disruptions.
• Service level agreement (SLA) impact: Databand shows data lineage and identifies downstream data pipelines affected by a data pipeline failure. If there is an SLA defined for data delivery, alerts help recognize and maintain SLA compliance.
• Performance and runtime anomalies: Databand monitors the duration of data pipeline runs and learns to detect anomalies, flagging them when necessary. Users do not need to be aware of the pipeline’s duration; Databand learns from its historical data.
• Status: Databand monitors the status of runs, including whether they are failed, canceled or successful.
• Data validation: Databand observes data value ranges over time and sends an alert upon detecting anomalies. This includes typical statistics such as mean, standard deviation, minimum, maximum and quartiles.

Users can set alerts by using the Databand user interface, which is uncomplicated and features an intuitive dashboard that monitors and supports workflows. It provides in-depth visibility through directed acyclic graphs, which is useful when dealing with many data pipelines. This all-in-one system empowers support teams to focus on areas that require attention, enabling them to accelerate deliverables.

IBM Enterprise Data’s mergers and acquisitions have enabled us to enhance our data pipelines with Databand, and we haven’t looked back. We are excited to offer you this transformative software that helps identify data incidents earlier, resolve them faster and deliver more reliable data to businesses.