Benchmarks¶

Triplet's benchmark suite exercises representative IIIF Image API requests across source formats, cache states, and concurrency levels. The full results are in fixtures/benchmark/README.md.

Benchmarks are most useful when the source and cache locations are explicit:

sources:
  default: file
  file:
    root: ./fixtures/benchmark/images

cache:
  root: ./deploy/compose/cache
  source_root: ./deploy/compose/source-cache

iiif:
  image:
    max_output_pixels: 100000000
    max_source_pixels: 250000000
    max_concurrent_transforms: 4

Summary¶

Dimension	Result
Success rate	100% across the benchmark request matrix.
Uncached latency	Median latency stays in the low single-digit second range for the tested transforms, with p99 latency remaining bounded under concurrent load.
Cached latency	Cached responses return substantially faster than uncached transforms because encoded derivatives are served from Triplet's cache layer.
JP2 support	JP2 source images complete successfully when the runtime libvips build includes OpenJPEG.
Concurrency	Triplet handles concurrent transform traffic with bounded latency growth across the tested concurrency levels.
CPU efficiency	Cached responses avoid repeat transform work; uncached responses keep CPU use proportional to requested image operations.
Memory footprint	Runtime memory remains bounded across source formats and concurrency levels in the benchmark matrix.
Output size	Derivative output sizes are tracked per request type so format and quality changes can be evaluated over time.

Running Benchmarks¶

Use the project Make targets to regenerate benchmark fixtures and run the IIIF benchmark suite:

make benchmark-fixtures
make benchmark-iiif

Pull request benchmark runs use:

make benchmark-iiif-pr