Definition
Response time (or latency) is the time between sending a request and receiving the response, usually measured in milliseconds. For monitoring, it's how long each check takes to complete.
Response time matters even when a service is technically "up." A page that takes 8 seconds to load is failing users in practice, and rising response times are often the first visible symptom of an overloaded system heading toward an outage.
Why It Matters
Slow is the new down. Users abandon slow pages, search engines penalize them, and APIs that respond slowly break the apps that depend on them. Tracking response time lets you catch performance regressions early — before they turn into timeouts and downtime.
How It Works
A monitoring tool times each check from request to response and records the value. Over many checks you get trends, averages, and percentiles. Setting a slow-response threshold lets you alert when response time crosses a level that signals trouble, even while checks still technically succeed.
Real-World Example
An API normally responds in about 180 ms. After a traffic spike, response times climb to 1,400 ms. The monitor's response-time chart shows the regression and a slow-response alert fires — the team scales the database before requests start timing out and causing a real outage.
Best Practices
- Set a slow-response threshold so degradation alerts before downtime
- Track response-time trends, not just the latest value
- Look at percentiles (p95/p99), not only the average
- Monitor response time on critical endpoints specifically
- Correlate spikes with deploys, traffic, and dependency issues
Common Mistakes
- Only checking up/down and ignoring how slow responses are
- Watching averages while tail latency quietly hurts users
- Having no threshold, so slow degradation goes unnoticed
- Measuring response time too infrequently to see trends
- Treating a fast homepage as proof the whole app is fast
In Monitoristic
Monitoristic records response time on every check and charts it over time, so you can spot regressions before they become outages. Pair it with a slow-response threshold to get alerted when an endpoint is up but degrading.