Next: , Previous: , Up: Background   [Contents][Index]

1.3 Latency and throughput

Most network benchmarks consider throughput, or the average amount of data being pushed around per unit of time. While important for batch applications (for example benchmarks), average throughput is mostly irrelevant when it comes to interactive web usage. What is more important is a transaction’s median latency, or whether the data starts to trickle down before the user gets annoyed.

Typical web caches optimise for throughput — for example, by consulting sibling caches before accessing a remote resource. By doing so, they significantly add to the median latency, and therefore to the average user frustration.

Polipo was designed to minimise latency.