Tail latency
The slow end of the distribution — p95, p99, p99.9.
Origin
Practitioner term, formalised by Dean & Barroso's "The Tail at Scale" (2013). The argument: for a fan-out request to 100 servers, system p99 ≈ each-server p50.
Where it shows up in production
- Google web search Tail-latency mitigation (hedged requests, tied requests) cuts p99 by half at modest cost.
- Netflix request hedging Documented in their tech blog as part of their resilience strategy.
On Semicolony
Sources & further reading
Found this useful?