.blog

Field notes from the lake.

Design rationale, format-deep-dives, and release notes. For day-to-day activity, follow @limnifs on GitHub.

• 2026-08-26

  LimniFS v0.3.0 — The Read Path Catches Up: 48 GiB of Wasted Decode, Gone by Construction

  Seekable drop containers in the format itself, SIEVE caches, O(1) slab lookups, and a public reader API. A cold 8 KiB window decodes one 256 KiB frame instead of the whole drop: 98× faster first reads, 83× faster cold windowed reads.

  #release#filesystem#performance

• 2026-08-22

  LimniFS v0.2.51 — On crates.io, Beating the Reference Brotli, and a 39× Create-Speed Week

  First crates.io releases via trusted publishing, CSV ratio 2.99% vs the C reference 6.23%, streaming writer, 0.2 s dedup images, offline Ed25519 signing, seven-platform binaries.

  #release#performance#supply-chain

• 2026-08-05

  LimniFS v0.2.36 — Pure-Rust Codec Stack, Read-Write Images, 18 Codecs

  Four weeks after v0.1.0: read-write images with crash-safe WAL, 18 codec variants in a tournament selector, 9 profiles, streaming write API, mmap-bounded peak memory, cross-file compress cache, and 5–8× speedups across the benchmark suite. Only one external codec dep remains (brotli).

  #release#performance#codecs

• 2026-08-01

  LimniFS v0.1.0 — A Pure-Rust Content-Addressed Filesystem

  The first release of LimniFS: a Layered, Immutable, Merkle-rooted, Network Image filesystem. Pure Rust, seven codecs, BLAKE3 content addressing, faster than DwarFS.

  #release#filesystem#rust