Latency is not one number

An in-memory lookup is usually far cheaper than a disk access; a network round trip adds more cost, and a cross-region round trip can take tens to hundreds of milliseconds. Exact numbers vary by hardware, load, payload, and distance, so memorize the scale rather than a false precision.

Every serial dependency adds to end-to-end latency. Three sequential 30 ms calls already consume about 90 ms before application work and queuing.