Benchmarks

This page mirrors the scenarios from the pnpm benchmarks page and adds aube to the comparison. The package manager versions, the test package set, and the scenario definitions are the same so the numbers can be read alongside pnpm's published results.

Methodology

Each row is a different starting state — what's already on disk before the command runs. This matters far more for install time than raw download speed, because most real-world installs have some state to reuse.

Scenarios

Scenario	cache	lockfile	node_modules
clean install	❌	❌	❌
with cache	✅	❌	❌
with lockfile	❌	✅	❌
with node_modules	❌	❌	✅
with cache, lockfile, node_modules	✅	✅	✅
with cache, lockfile	✅	✅	❌
with cache, node_modules	✅	❌	✅
with lockfile, node_modules	❌	✅	✅
update	✅	✅	✅

• clean install is the worst case — a fresh CI runner with no caches.
• with lockfile is the typical CI hit-rate scenario when restoring a lockfile but no node_modules cache.
• with cache, lockfile, node_modules is the typical local-dev no-op case.
• update runs <pm> update instead of <pm> install.

Real-world dependency set

A medium-sized package set spanning the major framework / tooling ecosystems (React, Vue, Express, ESLint, TypeScript, RxJS, …) — see bench/fixture/package.json. Lower is better. Numbers are from bench/run.sh, which uses hyperfine to time three runs of each command after restoring the relevant on-disk state.

npm Yarn pnpm aube

clean install

npm

4.18s

Yarn

4.59s

pnpm

1.02s

aube

824ms

with cache

npm

1.54s

Yarn

2.81s

pnpm

849ms

aube

187ms

with lockfile

npm

1.23s

Yarn

1.86s

pnpm

480ms

aube

329ms

with node_modules

npm

266ms

Yarn

4.21s

pnpm

—

aube

956ms

with cache, lockfile, node_modules (no-op)

npm

215ms

Yarn

139ms

pnpm

262ms

aube

43ms

with cache, lockfile

npm

919ms

Yarn

551ms

pnpm

457ms

aube

48ms

with cache, node_modules

npm

234ms

Yarn

2.89s

pnpm

465ms

aube

187ms

with lockfile, node_modules

npm

258ms

Yarn

143ms

pnpm

278ms

aube

1.02s

update

npm

863ms

Yarn

2.05s

pnpm

1.34s

aube

161ms

action	npm	Yarn	pnpm	aube
clean install	4.18s	4.59s	1.02s	824ms
with cache	1.54s	2.81s	849ms	187ms
with lockfile	1.23s	1.86s	480ms	329ms
with node_modules	266ms	4.21s	—	956ms
with cache, lockfile, node_modules (no-op)	215ms	139ms	262ms	43ms
with cache, lockfile	919ms	551ms	457ms	48ms
with cache, node_modules	234ms	2.89s	465ms	187ms
with lockfile, node_modules	258ms	143ms	278ms	1.02s
update	863ms	2.05s	1.34s	161ms

Reproduce locally

cargo build --release
AUBE_BIN=$PWD/target/release/aube bench/run.sh --runs 3

The script writes bench/results.json and a markdown table at bench/results.md.

The headline result is the no-op case (with cache, lockfile, node_modules): 43 ms, which is what every aube test invocation pays when nothing has changed since the last install. That's 6× faster than pnpm's 262 ms and an order of magnitude faster than npm and Yarn.

aube test vs pnpm install-test

A common workflow on a fresh checkout (or after pulling new commits) is "install dependencies, then run tests". pnpm exposes this as pnpm install-test — a single command that installs and then runs the test script. npm has the same shortcut as npm install-test (npm it); Yarn users typically chain yarn install && yarn test.

aube collapses this further. Because aube test (and any other aube run target) auto-installs when the lockfile or package.json is stale, you don't need a separate command at all — just aube test.

command	what it does	aube equivalent
npm install-test (npm it)	install, then npm test	aube test
yarn install && yarn test	install, then yarn test	aube test
pnpm install-test	install, then pnpm test	aube test

Numbers

Same fixture set as above, "install + test" measured end-to-end. The "test" script is a no-op so this measures the install pipeline plus the cost of shelling out to run the script.

action	npm install-test	yarn && test	pnpm install-test	aube test
clean	31.5s	7.6s	7.8s	TBD
with cache	13.6s	5.9s	4.8s	TBD
with cache, lockfile	4.6s	5.7s	2.4s	TBD
with cache, lockfile, node_modules (no-op install)	1.4s	5.3s	850ms	TBD

The last row is the most interesting one for day-to-day work: it's what happens every time you run your tests when nothing has changed. aube's goal here is to make the staleness check effectively free so aube test feels the same as just running the underlying test command directly.