ReferenceGlossary
Tail at scale
When one request fans out to many servers or waits for many replies, the slowest reply sets its latency, so a delay that is rare on each server becomes common for the whole request. With 100 servers each slow 1% of the time, 63% of requests meet at least one slow reply. Dean and Barroso described it in 2013, with fixes such as hedged requests.
Performance
In Quorum#
A request collects the tail events of every Node on its path: two independent chances a and b combine to 1 − (1 − a)(1 − b), so a long chain of Nodes that each stall now and then has a worse p99 than any one of them. A Wide-column Store read waits for the R-th fastest of N replicas, and at R = N it waits for the slowest one. Percentiles research shows the p99 where this appears.