Infrastructure performance indicators
Inspect infrastructure performance indicators
The four classic resources — CPU, memory, disk and network — plus the queues in front of them.
| Indicator | Watch for |
|---|---|
| CPU | Sustained saturation; also low CPU with high latency, which means waiting, not computing |
| Memory | Growth without plateau — a leak; and paging, which destroys latency |
| Disk | IOPS and queue depth; latency rises sharply near the limit |
| Network | Throughput, retransmits, connection exhaustion |
Utilisation, saturation, errors
For each resource ask three questions: how utilised is it, how much work is queued for it, and is it producing errors? Utilisation alone misleads — a disk at 60% utilisation with a deep queue is already the bottleneck, because queueing begins well before saturation.
The most useful negative signal
Low CPU with high latency is the shape worth recognising: the service is not computing, it is waiting — on a dependency, a lock, a disk or a connection pool. Teams reading only CPU conclude the host is healthy and go looking in the wrong place.
Infrastructure explains, users decide
These indicators are diagnostic. Alert on the user-facing signals — availability, latency, error rate — and use CPU, memory, disk and network to explain why. That ordering is what keeps 3am pages meaningful.
Primary sources