Benchmarks
The benchmark environment and results for asyncmux. Throughput per scenario, with charts and tables.
Environment
| Item | Value |
|---|---|
| CPU | 12th Gen Intel(R) Core(TM) i9-12900K |
| OS | Linux x64 |
| Node.js | v24.15.0 |
| Vitest | 4.1.11 |
| Date | 2026-09-26 |
Measurements used packages/asyncmux/benchmarks/asyncmux.bench.ts, run with the following command.
npx vitest bench --config ./.config/vitest.bench.ts --runThroughput
The horizontal axis shows lock acquisitions per second. Scenarios where a single run performs multiple lock operations are normalized by the number of lock operations per run. The “(n)” in the chart corresponds to the “Operation” column in Operations and results.
Operations and results
The table lists the median hz (runs per second) of 3 measurements. “Lock acquisitions/sec” is normalized by multiplying by the number of lock operations per run. The two abort scenarios never acquire a lock and are excluded from the chart.
| Operation | Description | Lock ops/run | Runs/sec | Lock acquisitions/sec |
|---|---|---|---|---|
| (1) | lock / release (no key) | 1 | 1,146,546 | 1,146,546 |
| (2) | lock / release (key) | 1 | 998,446 | 998,446 |
| (3) | rLock / release | 1 | 1,144,467 | 1,144,467 |
| (4) | rLock / release (key) | 1 | 835,703 | 835,703 |
| (5) | 2 workers x 4 ops (same key) | 8 | 73,622 | 588,972 |
| (6) | 2 workers x 16 ops (same key) | 32 | 22,983 | 735,462 |
| (7) | 32 concurrent requests (no key) | 32 | 21,211 | 678,742 |
| (8) | 128 concurrent requests (same key) | 128 | 2,157 | 276,045 |
| (9) | sequential x 16 | 16 | 56,861 | 909,773 |
| (10) | sequential x 64 | 64 | 15,564 | 996,121 |
| (11) | drain 50 queued R/W requests | 50 | 14,424 | 721,185 |
| (12) | drain 200 queued R/W requests | 200 | 3,164 | 632,720 |
| (13) | double release (throws LockReleasedError) | 1 | 215,682 | 215,682 |
| (14) | reader-writer cache access (12R + 3W + global W) | 16 | 43,757 | 700,108 |
| — | immediately aborted signal (rejects) | — | 152,145 | — |
| — | abort while waiting in queue | — | 96,851 | — |
Observations
- Uncontended lock acquisition and release runs at about 1M operations per second.
- Read lock throughput is on par with write locks.
- With 128 concurrent requests on the same key, throughput drops to about 280K/sec while preserving mutual exclusion and FIFO.
- Error paths such as double release and aborts also handle 90K+ operations per second.
Re-running
Benchmarks are defined in packages/asyncmux/benchmarks/asyncmux.bench.ts. To measure on your own machine, run the following command in packages/asyncmux.
npx vitest bench --config ./.config/vitest.bench.ts --run