Changelog
Probe agents from version 2.4 measure HTTP response times more precisely. Every check runs on its own connection, so the value reflects your service alone. Typical readings are up to four times lower and far more consistent.
With agent versions 2.4 to 2.6 we reworked how probes run HTTP checks. Connection capacity scales with each probe, check starts are spread evenly across the interval, and every check opens a fresh connection.
A response time now covers exactly what a first-time visitor of your site experiences: the DNS lookup, the TCP and TLS handshake and the answer of your server. The new agents rolled out across the whole probe network from 10 October.
Across the same 102 HTTP monitors, comparing 5 to 9 October with 25 to 31 October:
91 of these 102 monitors now read more than 20 % lower. Readings are also more consistent: the 95th percentile used to be about ten times the median, now it is about six times. Ping and TCP checks are not affected by this change.
Response time charts show a step around 10 and 11 October, because values before and after that date were measured with different agent versions. If you set a response time alert based on earlier values, check whether it is still tight enough for your service, see Monitors in the docs.