leanstore · research

LeanStore

LeanStore is a research storage engine that manages database pages across RAM and SSDs while reducing buffer-management and synchronization overhead.

2018 design: Pointer swizzling replaces a page identifier with a memory pointer on the hot path. Each page has one owning reference; parents with swizzled children cannot be evicted. [1]

Cooling: An unswizzled page can remain in RAM in a FIFO queue. A new access can restore its pointer without reading storage. Before reuse, the minimum active thread epoch must exceed the page's unswizzling epoch; waiting in the queue alone is insufficient. [1]

Later design: The 2023 vmcache uses hardware page translation with explicit database-controlled reads and eviction, not ordinary file-backed mmap. Its basic form uses standard OS primitives; exmap is a separate Linux extension for scalable page-table manipulation. [2]

Version scope: The implementation repository distinguishes pointer-swizzling branches from VMCache branches. Neither the 2018 replacement protocol nor a later feature list describes every revision. [3]

Relations

related
B+ tree, CLOCK cache replacement, LLAMA

Citations

  1. Leis, Haubenschild, Kemper and Neumann. LeanStore. ICDE 2018, pp. 185-196; sections III-IV and V-A. DOI 10.1109/ICDE.2018.00026. (accessed 2026-11-15)
  2. Leis, Alhomssi, Ziegler, Loeck and Dietrich. Virtual-Memory Assisted Buffer Management. SIGMOD 2023; sections 3-4. (accessed 2026-11-15)
  3. LeanStore official README at fcb64bf39cf69c435173f76e062cbb305a9a998f: Cite, VMCache and Pointer Swizzling versions. (accessed 2026-11-15)

Further reading

  1. Technical University of Munich: LeanStore ICDE 2018 publication metadata, authors, pages and DOI. (accessed 2026-11-15)
  2. Hesela: B-epsilon trees and the trade-offs of buffered index updates. (accessed 2026-11-15)
Sources & caveats
  • Source synthesis, not a native LeanStore execution, crash test or reproduced benchmark.
  • The 2018 storage-manager comparison disabled transactions, logging, compaction and compression in the comparison engines; its TPC-C workload was not a full transactional-throughput benchmark. [1]
  • The pinned README separates feature branches. It is evidence of version organization, not verification of recovery behavior. [3]

JSON: /data/terms/leanstore.json

Published by Hesela team · Updated

Report a correction · Editorial standards · Reuse guidance

Cite this page
Hesela team. LeanStore. Hesela. 2026-11-15. https://hesela.dev/terms/leanstore/