The latency of a system between two events is the time from the first event (a packet arriving, a function being called) to the second (an order leaving, the function returning), measured for one occurrence. It is a random variable: each message has its own.
Ejemplos
Example 1.11 (Median against tail in a race)
Our latency has a lognormal body with median and a 2% chance of a stall of on average; the competitor’s median is with stalls of the same length 0.5% of the time. Simulated over a million races, we win 66.1%. Halving our median wins 98.0%. Halving our stall rate, which cuts our p99 from to , wins 66.7%: a large improvement of a percentile that races barely see (Figure 1.4).