Latency is the delay between making a request and receiving a response. For an online service, it affects whether a response arrives soon enough to be useful. A system that answers after thirty minutes may be available under a definition that requires only a non-error response, but it will feel unavailable to its users. section 6.12 section 6.2
In distributed systems, latency is also a design trade-off. PACELC asks what to prioritize when there is no network partition: low latency or consistency. Here, latency can be expressed as an upper bound for a non-error response, such as 500 ms. This adds a responsiveness question that CAP does not explicitly address. section 6.3 section 6.10.1
Small delays add up
Users can notice differences measured in milliseconds. Reported examples include a 1% drop in Amazon sales after a 100 ms increase in website load time and a 20% drop in Google search traffic after a 500 ms increase in result time. The practical question is not just whether a request succeeds, but how long users will wait. section 6.12.1
Three approaches address latency from the user’s perspective:
- Make responses faster by optimizing a critical code path, adding caching, improving database speed, or using a CDN. section 6.12.2
- Anticipate requests by preloading content. Netflix, for example, preloads part of a video while someone browses, reducing the delay they perceive when playback starts. section 6.12.2
- Make a wait feel shorter when the work cannot be accelerated. A progress bar may change the experience of waiting, but it does not make the underlying task finish sooner. section 6.12.2


































