system-design-primer
reference — fundamentals

Latency vs throughput

in short — written for this site
Latency is the time to perform an action. Throughput is the number of actions per unit of time.
latency
Time to produce a result, measured at the tail (p95, p99) rather than the mean — the average hides the requests that lose users.
throughput
Results per unit of time. Bounded by the slowest stage in the pipeline, never by the fastest.
from the primer — full sectionsynced

Latency is the time to perform some action or to produce some result.

Throughput is the number of such actions or results per unit of time.

Generally, you should aim for maximal throughput with acceptable latency.

Source(s) and further reading

sourcedonnemartin/system-design-primer / README.mdanchor#latency-vs-throughputsyncedsynced from donnemartin/system-design-primer@master · 2026-08-16