tars · research · also Timeliness-aware Adaptive Replica Selection

Tars replica selection

Tars is a replica-selection scheme that adapts ranking and rate control to the timeliness of server feedback.

The client tracks time since feedback receipt separately from transport delay. Ranking changes after a 100 ms receive-to-use interval; this is a design threshold, not a universal expiry time. The stale branch still uses stored quantities, so it is not wholly feedback-free. Server completion rate is measured separately from individual service duration. [1]

The arXiv evaluation uses simulated Poisson arrivals, exponential service and four parallel service slots by default, with 600,000 requests per run and five seeds. Its comparisons do not establish a production speedup. [1]

Relations

related
Adaptive replica selection, Feedback age, Tail latency

Citations

  1. Jiang et al. Tars: Timeliness-aware Adaptive Replica Selection for Key-Value Stores. arXiv:1702.08172v1 (2017), sections III-V. (accessed 2026-11-21)
Sources & caveats
  • Reviewed arXiv v1, sections III-V and Algorithms 1-2; no implementation reproduction claimed.

JSON: /data/terms/tars.json

Published by Hesela team · Updated

Report a correction · Editorial standards · Reuse guidance

Cite this page
Hesela team. Tars replica selection. Hesela. 2026-11-21. https://hesela.dev/terms/tars/