What does monitoring actually measure, across the three layers it covers?

Availability (is the server, service or URL alive), performance (CPU, RAM, storage, network throughput, latency) and application behaviour (response times, error rates, transactions).

• Availability — the binary question, answered with heartbeats, pings, port checks and URL probes. Cheap, coarse, and the first thing anyone builds.
• Performance — the resource layer: CPU, memory, disk, network throughput and latency. This is where you see why a service is degrading before it disappears.
• Applications — what the user actually experiences: response times, error rates, transactions completed or failed. A host can be perfectly healthy on both layers above while the application in it returns HTTP 500 to everyone.

The layers are ordered by how close they sit to the business. A green dashboard at the availability layer with an angry customer on the phone is the classic symptom of monitoring that never climbed past layer one — which is precisely the gap the Golden Signals and the business-first approach are meant to close.

Go deeper:

• Brendan Gregg — The USE Method — utilisation, saturation, errors for every resource: the systematic checklist for the performance layer.