SignalFx Inc., a startup that claims to have a radical new approach to system monitoring in the age of containers and micro services, is enhancing its core platform with new features to better detect anomalies in complex infrastructure.

The venture-backed company, which emerged from stealth about a year ago, is applying big data analytics principles to systems management, saying that a system-wide view of applications and infrastructure is the only way to understand what’s really going on in today’s multi-faceted environments.

SignalFx uses data collected from thousands of sources to create aggregations like percentiles, moving averages and growth rates in near-real-time. That enables administrators to set alerts based upon anomalies rather than individual faults, which were a luxury of relatively simple legacy computing environments that didn’t have to deal with a constant influx of clients, virtual machines, patches and application changes.

“New applications don’t look like yesterday’s,” said CEO Karthik Rau. “Element managers give you visibility into one component of your stack, but today’s apps are so complex that you need to be able to look across them, build analytics across populations and alert on those patterns very quickly.”

The approach taken by SignalFx, called SignalFlow, grew out of that pioneered by Facebook, where SignalFx co-founder Phillip Liu built much of the social network’s original application monitoring systems. Instead of monitoring individual components, the service gathers data continually across a variety of metrics into a centralized analytics engine that can identify patterns that are meaningful across a range of services.

Rather than pinpointing a single fault – or even classifying an anomaly as a problem – “We’re identifying potential causes and helping customers triage them,” Rau said. Here are the new features being introduced today:

• A one-screen snapshot of infrastructure health via a navigator combines a high-level view with the ability to drill down by dimensions such as application, region, service and cluster to find the hosts, virtual machines, containers, processes, metrics, outliers and alerts for that grouping. This enables quick isolation of hot spots or outliers at any level of the stack, the company said.
• A new outlier detection feature proactively identifies abnormal performance patterns and alerts administrators through the means of their choice. The feature enhances users’ existing ability to build their own detectors using SignalFlow with pre-packaged analytics that identify and alert administrators to outliers. Detectors are configurable and can be based on any metric to surface outliers from a population or historically over time. For example, a user could define a latency metric and get an alert if a significant number of outliers are out of range or deviate from historical metrics at any point, Rau said.
• Built-in detectors make it easier to set up better alerts. SignalFx provides users with a starting point for good alert design–embedding the complex statistical methods and adaptive thresholds needed to reduce alert noise. The detectors then provide pre-packaged and customizable alert configurations as templates for all the platforms and technologies supported by the monitoring software. Relevant built-in and user-created detectors are surfaced in any view for easy tuning, activation and subscription, the company said.

Pricing is based on the data ingestion rate. Rau said the company advises clients to budget about $15 per server per month to get a satisfactory monitoring level.

Image courtesy SignalFx