The RAIL* model for user-perceived performance does a really nice job of distilling rules of thumb for latency into a handful of specific metrics / perf targets.

*https://developers.google.com/web/fundamentals/performance/r...