<!-- llms-explorer concept facts · https://llms-explorer.com/tree/tail-latency-mitigation/ · pack 2026-10-15 · ~447 tokens -->

# Tail Latency Mitigation

> Depth-first rabbithole dossier for Tail Latency Mitigation; source-anchored research pack.

Parent: [Client Side Operations Timeout (CSOT)](https://llms-explorer.com/tree/client-side-operations-timeout-csot/) · 3 facets · 3 facts · page: https://llms-explorer.com/tree/tail-latency-mitigation/

## Measurements and reference values

- [low] For a real Google system with 10ms latency at the 99% percentile for any single request, the 99% percentile for all requests to finish is 140ms, and the 95% percentile is 70ms. Sources: https://blog.acolyer.org/2015/01/15/the-tail-at-scale/ — source: `~/.global-ai-hub/research/.pipeline/mongodb-2026-10-14/frb/tail-latency-mitigation-bd860579dd/rabbithole-synthesis.md#core`

## Problems, failure modes and limitations

- [low] TTFT-calibrated hedging achieves comparable tail improvement at roughly 19.8 percent overhead and only fires on the slow twenty percent, the actual cache misses, leaving normal requests untouched. Sources: https://infoq.com/articles/adaptive-hedged-requests-p99-latency — source: `~/.global-ai-hub/research/.pipeline/mongodb-2026-10-14/frb/tail-latency-mitigation-bd860579dd/rabbithole-synthesis.md#core`

## Facts and statements

- [low] Static hedge thresholds appear effective in benchmarks but break in production as latency distributions shift with load, deployments, and time of day, requiring continuous manual tuning that rarely happens in practice. Sources: https://infoq.com/articles/adaptive-hedged-requests-p99-latency — source: `~/.global-ai-hub/research/.pipeline/mongodb-2026-10-14/frb/tail-latency-mitigation-bd860579dd/rabbithole-synthesis.md#antipatterns`

## Related concepts

- Latency — is a part of Tail Latency Mitigation
- Tail — is a part of Tail Latency Mitigation
