`deno bench`, benchmarking tool

Deno has a built-in benchmark runner that you can use for checking performance of JavaScript or TypeScript code.

Quickstart

Firstly, let's create a file url_bench.ts and register a bench using the Deno.bench() function.

// url_bench.ts
Deno.bench("URL parsing", () => {
  new URL("https://deno.land");
});

Secondly, run the benchmark using the deno bench subcommand.

deno bench url_bench.ts
cpu: Apple M1 Max
runtime: deno 1.21.0 (aarch64-apple-darwin)

file:///dev/deno/url_bench.ts
benchmark        time (avg)             (min … max)       p75       p99      p995
--------------------------------------------------- -----------------------------
URL parsing   17.29 µs/iter  (16.67 µs … 153.62 µs)  17.25 µs  18.92 µs  22.25 µs

Writing benchmarks

To define a benchmark you need to register it with a call to the Deno.bench API. There are multiple overloads of this API to allow for the greatest flexibility and easy switching between the forms (eg. when you need to quickly focus a single bench for debugging, using the only: true option):

// Compact form: name and function
Deno.bench("hello world #1", () => {
  new URL("https://deno.land");
});

// Compact form: named function.
Deno.bench(function helloWorld3() {
  new URL("https://deno.land");
});

// Longer form: bench definition.
Deno.bench({
  name: "hello world #2",
  fn: () => {
    new URL("https://deno.land");
  },
});

// Similar to compact form, with additional configuration as a second argument.
Deno.bench("hello world #4", { permissions: { read: true } }, () => {
  new URL("https://deno.land");
});

// Similar to longer form, with bench function as a second argument.
Deno.bench(
  { name: "hello world #5", permissions: { read: true } },
  () => {
    new URL("https://deno.land");
  },
);

// Similar to longer form, with a named bench function as a second argument.
Deno.bench({ permissions: { read: true } }, function helloWorld6() {
  new URL("https://deno.land");
});

Async functions

You can also bench asynchronous code by passing a bench function that returns a promise. For this you can use the async keyword when defining a function:

Deno.bench("async hello world", async () => {
  await 1;
});

Critical sections

Sometimes the benchmark case needs to include setup and teardown code that would taint the benchmark results. For example, if you want to measure how long it takes to read a small file, you need to open the file, read it, and then close it. If the file is small enough the time it takes to open and close the file might outweigh the time it takes to read the file itself.

To help with such situations you can Deno.BenchContext.start and Deno.BenchContext.end to tell the benchmarking tool about the critical section you want to measure. Everything outside of the section between these two calls will be excluded from the measurement.

import { readAll } from "https://deno.land/std@0.221.0/io/read_all.ts";

Deno.bench("foo", async (b) => {
  // Open a file that we will act upon.
  const file = await Deno.open("a_big_data_file.txt");

  // Tell the benchmarking tool that this is the only section you want
  // to measure.
  b.start();

  // Now let's measure how long it takes to read all of the data from the file.
  await readAll(file);

  // End measurement here.
  b.end();

  // Now we can perform some potentially time-consuming teardown that will not
  // taint out benchmark results.
  file.close();
});

Grouping and baselines

When registering a bench case, it can be assigned to a group, using Deno.BenchDefinition.group option:

// url_bench.ts
Deno.bench("url parse", { group: "url" }, () => {
  new URL("https://deno.land");
});

It is useful to assign several cases to a single group and compare how they perform against a "baseline" case.

In this example we'll check how performant is Date.now() compared to performance.now(), to do that we'll mark the first case as a "baseline" using Deno.BenchDefinition.baseline option:

// time_bench.ts
Deno.bench("Date.now()", { group: "timing", baseline: true }, () => {
  Date.now();
});

Deno.bench("performance.now()", { group: "timing" }, () => {
  performance.now();
});

$ deno bench time_bench.ts
cpu: Apple M1 Max
runtime: deno 1.21.0 (aarch64-apple-darwin)

file:///dev/deno/time_bench.ts
benchmark              time (avg)             (min … max)       p75       p99      p995
--------------------------------------------------------- -----------------------------
Date.now()         125.24 ns/iter (118.98 ns … 559.95 ns) 123.62 ns 150.69 ns 156.63 ns
performance.now()    2.67 µs/iter     (2.64 µs … 2.82 µs)   2.67 µs   2.82 µs   2.82 µs

summary
  Date.now()
   21.29x times faster than performance.now()

You can specify multiple groups in the same file.

Running benchmarks

To run a benchmark, call deno bench with the file that contains your bench function. You can also omit the file name, in which case all benchmarks in the current directory (recursively) that match the glob {*_,*.,}bench.{ts, tsx, mts, js, mjs, jsx} will be run. If you pass a directory, all files in the directory that match this glob will be run.

The glob expands to:

files named bench.{ts, tsx, mts, js, mjs, jsx},
or files ending with .bench.{ts, tsx, mts, js, mjs, jsx},
or files ending with _bench.{ts, tsx, mts, js, mjs, jsx}

# Run all benches in the current directory and all sub-directories
deno bench

# Run all benches in the util directory
deno bench util/

# Run just my_bench.ts
deno bench my_bench.ts

⚠️ If you want to pass additional CLI arguments to the bench files use -- to inform Deno that remaining arguments are scripts arguments.

# Pass additional arguments to the bench file
deno bench my_bench.ts -- -e --foo --bar

deno bench uses the same permission model as deno run and therefore will require, for example, --allow-write to write to the file system during benching.

To see all runtime options with deno bench, you can reference the command line help:

deno help bench

Filtering

There are a number of options to filter the benches you are running.

Command line filtering

Benches can be run individually or in groups using the command line --filter option.

The filter flags accept a string or a pattern as value.

Assuming the following benches:

Deno.bench({
  name: "my-bench",
  fn: () => {/* bench function zero */},
});
Deno.bench({
  name: "bench-1",
  fn: () => {/* bench function one */},
});
Deno.bench({
  name: "bench2",
  fn: () => {/* bench function two */},
});

This command will run all of these benches because they all contain the word "bench".

deno bench --filter "bench" benchmarks/

On the flip side, the following command uses a pattern and will run the second and third benchmarks.

deno bench --filter "/bench-*\d/" benchmarks/

To let Deno know that you want to use a pattern, wrap your filter with forward-slashes like the JavaScript syntactic sugar for a regex.

Bench definition filtering

Within the benches themselves, you have two options for filtering.

Filtering out (ignoring these benches)

Sometimes you want to ignore benches based on some sort of condition (for example you only want a benchmark to run on Windows). For this you can use the ignore boolean in the bench definition. If it is set to true the bench will be skipped.

Deno.bench({
  name: "bench windows feature",
  ignore: Deno.build.os !== "windows",
  fn() {
    // do windows feature
  },
});

Filtering in (only run these benches)

Sometimes you may be in the middle of a performance problem within a large bench class and you would like to focus on just that single bench and ignore the rest for now. For this you can use the only option to tell the benchmark harness to only run benches with this set to true. Multiple benches can set this option. While the benchmark run will report on the success or failure of each bench, the overall benchmark run will always fail if any bench is flagged with only, as this is a temporary measure only which disables nearly all of your benchmarks.

Deno.bench({
  name: "Focus on this bench only",
  only: true,
  fn() {
    // bench complicated stuff
  },
});

JSON output

To retrieve the output as JSON, use the --json flag:

$ deno bench --json bench_me.js
{
  "runtime": "Deno/1.31.0 x86_64-apple-darwin",
  "cpu": "Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz",
  "benches": [
    "origin": "file:///dev/bench_me.js",
    "group": null,
    "name": "Deno.UnsafePointerView#getUint32",
    "baseline": false,
    "result": {
      "ok": {
        "n": 49,
        "min": 1251.9348,
        "max": 1441.2696,
        "avg": 1308.7523755102038,
        "p75": 1324.1055,
        "p99": 1441.2696,
        "p995": 1441.2696,
        "p999": 1441.2696
      }
    }
  ]
}

deno bench, benchmarking tool

Quickstart​

Writing benchmarks​

Async functions​

Critical sections​

Grouping and baselines​

Running benchmarks​

Filtering​

Command line filtering​

Bench definition filtering​

Filtering out (ignoring these benches)​

Filtering in (only run these benches)​

JSON output​