How to write benchmarks

  • Updated: 2018 August 9

Overview

This guide will walk you through the process of writing a benchmark, running it at every commit, and automatically tracking the results in the Performance Dashboard.

Today we support automating benchmarks for these projects: * Garnet (Also runs Zircon benchmarks) * Peridot * Topaz

Writing a benchmark

Fuchsia benchmarks are command-line executables that produce a JSON results file. The executable must meet the following criteria:

  1. It accepts the location to the results file as a command line flag.
  2. It produces JSON results that match the benchmark results schema:

Building your benchmark

Your benchmark executable should be built into a Fuchsia package. For more information please read the Fuchsia Build overview.

Automating your benchmark

We have shell scripts that run all of a layer's benchmarks at every commit to that layer.

These shell scripts are written using a helper library called benchmarking. Add a command to the appropriate script to execute your test. See the existing commands for examples.

Testing

At this point, you're ready to build Fuchsia and test that your benchmark runs successfully. Run the following in a shell:

jiri update -gc
# Benchmarks are not included in production packages, so use $layer/packages/kitchen_sink
# or they will not be built.
fx set core.<board> --with //bundles:kitchen_sink
fx build && fx run

Once the Fuchsia shell is loaded:

# Run just your benchmark
run my_benchmark [options]

# Run all benchmarks for $layer
/pkgfs/packages/${layer}_benchmarks/0/bin/benchmarks.sh /tmp

If no errors occurred, you should see your benchmark's output file in /tmp, along with the results files of other benchmarks.

Tracking in the performance dashboard

Please see the Performance Dashboard User Guide