Build a latency budget
A short round trip is one part of a fast connection. Measure the steps that surround it.
Follow one request
A connection can spend time resolving a name, establishing transport, negotiating encryption and waiting for application work. A single ping reports none of these stages in full. Start with one destination and keep the connection conditions stable while taking measurements.
Separate connection setup from reuse
The first request may include DNS and a new connection. A later request can reuse both. Record cold and warm results separately rather than averaging them into a number that describes neither experience.
Estimate a first request
An illustrative model, not a live test. Transport versions, reuse and parallel work can change the result.
Estimated total:
Change the largest measured cost
A closer endpoint may improve network time while leaving application latency untouched. A faster resolver may do little for a connection that is already warm. Tie each proposed change to the stage it can influence, then repeat the same measurement.