Skip to content

performance-benchmarks.md

Latest commit

 

History

History
67 lines (44 loc) · 2.15 KB

performance-benchmarks.md

File metadata and controls

67 lines (44 loc) · 2.15 KB

Performance Benchmarks

PlainNAS includes a built-in benchmark command to help evaluate performance and to make results easy to share with other developers.

What it measures

plainnas bench runs three practical benchmarks:

  • Index build: scans a dataset and builds the on-disk search index (and Pebble metadata).
  • Search queries: executes repeated SearchIndex calls and reports latency (avg/p50/p95).
  • Thumbnail generation: generates WEBP thumbnails for many images (through the same vipsthumbnail path used in production).

It also prints RSS (process resident memory) and Go heap before/after each phase.

Requirements

  • Linux (PlainNAS is Linux-only).
  • For thumbnail benchmarks: vipsthumbnail must be installed.
    • On Debian/Ubuntu: sudo apt install libvips-tools

If vipsthumbnail is not available, the thumbnail phase is skipped (and the command prints a warning).

Run

Quick run (auto-generated dataset)

sudo go run ./main.go bench

Notes:

  • By default the command generates a temporary dataset and uses a temporary data directory, so it will not touch /var/lib/plainnas.
  • Use --keep to keep the temp dirs so you can inspect index sizes.

Run on a real dataset

sudo go run ./main.go bench --dataset /mnt/storage

JSON output (for pasting into issues/PRs)

sudo go run ./main.go bench --json > bench.json

Useful flags

  • --files, --images: control generated dataset size.
  • --thumb-size, --thumb-quality, --thumb-concurrency: control thumbnail workload.
  • --queries, --limit: control search workload.
  • --show-hidden: include dotfiles in indexing.

Interpreting results

The key numbers for presenting value are typically:

  • Index build ops/s: how fast the index builds for large datasets.
  • Search p95: tail latency for interactive search.
  • Thumbnails ops/s: throughput for generating image thumbnails.

When comparing against other projects, try to keep the same:

  • Dataset (file count, file name patterns, image sizes)
  • Storage hardware (SSD/HDD)
  • CPU / memory configuration
  • vipsthumbnail version (thumbnail performance depends on libvips)