triad · research · also TRIAD LSM store

TRIAD

TRIAD is a research key-value store that combines hot-key retention, deferred compaction and commit-log reuse to reduce LSM maintenance writes.

Mechanism: TRIAD-LOG pairs an existing commit log with a sorted latest-offset index instead of rewriting its values during L0 flush. Retained hot keys still need logging. [1]

Accounting: the paper's flush-normalized ratio is not total engine output divided by application input. [1] See database-write-amplification for the denominator distinction.

Durability: the author README lists --disable_wal=false, while the pinned WriteOptions separately defaults sync to false. WAL presence alone does not establish synchronized acknowledgment. These source defaults do not establish the settings of every reported experiment. [2] [3]

Relations

related
Log-structured merge-tree, Memtable, Write-ahead log, LSM compaction, Database write amplification

Citations

  1. Balmau et al. TRIAD. USENIX ATC 2017, pp. 363-375; sections 4.1-4.3 and 5.1-5.3. (accessed 2026-11-10)
  2. TRIAD author prototype, pinned README: benchmark controls and enabled WAL. (accessed 2026-11-10)
  3. TRIAD author prototype, pinned WriteOptions: synchronization and WAL defaults. (accessed 2026-11-10)

Further reading

  1. Hesela: TRIAD log reuse and flush-normalized accounting. (accessed 2026-11-10)
Sources & caveats
  • Primary-source review of the full ATC 2017 paper and pinned author repository. No TRIAD execution, crash test or benchmark reproduction.

JSON: /data/terms/triad.json

Published by Hesela team · Updated

Report a correction · Editorial standards · Reuse guidance

Cite this page
Hesela team. TRIAD. Hesela. 2026-11-10. https://hesela.dev/terms/triad/