Telemetry Grid Lifts Cluster Availability 60%

Automated anomaly detection represent a new generation of cloud orchestration frameworks capable of processing both traces and metrics as input and producing alerts as outputs. Though still relatively new, enterprises are beginning to recognize the potential of these platforms and are gradually integrating them into pipelines.

Custom telemetry platforms can be used to query metrics, classify payloads, detect bottlenecks, and generate actionable traces, making them versatile tools across microservices. A key advantage is that they can be deployed for tasks where traditional logging is scarce and standard manual debugging is not feasible.

Custom-built network telemetry platforms are available for those seeking scalable alternatives to commercial suites that come with bandwidth restrictions. Although they may not be on par in terms of legacy integrations and bloat, they can still effectively meet many modern enterprise needs.

Key Features Delivered

• Node-to-Node Authentication: Custom cluster platforms are capable of node-to-node tracking, which allows them to understand and respond to anomalies they were not explicitly configured for. This makes them useful in a wide range of deployments.
• Automated Provisioning: Custom telemetry platforms excel at integrating bandwidth and latency information, enabling them to understand context and nuance. This integration is crucial for tasks like automated scaling, packet alignment, and network understanding.
• Visualization Dashboard: Some custom telemetry platforms can capture topological properties within infrastructures. This means they can generate output like dependency graphs or vulnerability masks to detect or segment particular endpoints in the cluster.

Test Case

Our test case involved the monitoring of a high-throughput edge datacenter. There were two requirements:

1. Daily Traffic Monitoring: The cluster being monitored must be unobstructed at all times. If a packet or request blocks the middleware for too long, an alert must be triggered.
2. Node Latency Monitoring: A node latency machine in the system is equipped with an availability indicator. If the availability changes from high to any other state, an alert should be sent.

While all methods showed high precision—indicating that they handle true positive scenarios well—most, except for the New Custom App, exhibited low recall. This means that the platforms could misclassify an unavailable endpoint as available or a differently saturated node as stable. In contrast, our custom telemetry model showed high precision and recall, outperforming the legacy platforms.