When users say the server is slow, increasing instance size is a tempting first move. It may help briefly, but a DNS delay, full connection pool, external API or heavy browser task remains untouched. Performance work begins by locating the wait.

Find who is slow and when

Compare region, device, account size, data shape and the time the issue began. Distinguish first paint from the completion of the action the user attempted.

Split the request path

Measure DNS, TCP and TLS, edge, application, database, external service, transfer and rendering. Correlate browser and server records with the same request ID.

Inspect the tail

P50, P95 and P99 by endpoint and data condition reveal users hidden by an average. Include failed requests that waited until timeout.

Reproduce before changing

Use production-like data sizes, network limits and devices. Compare the same scenario before and after a change, then check whether another segment became the new bottleneck.