Runs one or multiple benchmark tests
npm install -g bench-trialbench-trial benchmarks/map-helper-vs-entity.js -i 5Runs one (or more) BenchmarkJs test multiple times enough to get less ambiguous results, includes basic testing to make sure tests are reliable.
While running benchmarkjs to compare different versions of code I found out a couple of things:
- Ambiguous results: I noticed that the same benchmark tests were returning different results every time they executed. If they were re-run consecutively, I would get more operations per second on each benchmark. I believe the reason may be related to the v8 engine warming up and optimizing the code the more it ran, since if I let some time to "cool off" the operations per second for each test would decrease. These ambiguous results meant having to repeat tests to ensure some consistency.
- Unreliable execution: Occasionally I made changes to the benchmarked code and would overlook that it was not executing correctly, further compounding the issue of making the results unreliable.
-
Consistency: By running benchmark tests more than once, we can get median and average results and get a bigger picture with less fluctuation. Because the tests will run multiple times in succession, the code will get optimized by the engine, and we can use the median time as a more consistent and stable metric.
-
Reliable execution: By running simple assertion tests on each suite before the actual benchmark runs, we can be sure our tests are executing correctly.
bench-trial <file> [-i <iterations>] [-s]-i --iterations <iteration>iterations default to 10 iterations if not provided.-s --skip-testsif provided, it will skip the assertion tests.
The file provided to bench-trial should export an array of test suites, each test suite is an object in the form of:
{
name: string,
test: function,
benchmark: function
}
| Property | Type | Description |
|---|---|---|
| name | String |
Name that describes the test you are running |
| test | function |
function to run assertion test against the result of the code you want to benchmark |
| benchmark | function |
function to pass to benchmarkjs Suite that actually runs the benchmark |
- Synchronous methods are simple methods that expect a return value.
- Asynchronous methods are a bit different to benchmarkjs async methods, bench-trial expects async methods to follow the error-first callbacks.
bench-trial provides a convenience method that accepts the function to execute and a value to check against the result of the code you are testing. It takes care of async vs async depending on how you set the async flag.
test(test:function, value:*)To write your manual test see the manual test example below
- Test synchronous code example
- Test asynchronous code example
- Write manual test sync/asynchronous code example
This tool is a wrapper of benchmarkjs, so all credit related to benchmarking itself really goes to them.
Thanks to Paul Molluzzo for coming up with the name bench-trial!
Please read CONTRIBUTING.md for details on our code of conduct, and the process for submitting pull requests to us.
This project is licensed under the Apache License Version 2.0 - see the LICENSE file for details