By default, Telemetry runs all benchmarks in a fresh, clean profile. This isn't always desirable for benchmarking. To run with other types of profiles, use either --profile-type or --profile-dir.
- clean (default behavior) -- a clean, new profile. Note that benchmarks may include initialization routines that only happen on the first run.
- default -- the default profile associated with the browser. As far as Telemetry is concerned, this is a non-deterministic state.
- typical_user -- a pre-recorded profile that represents a typical user's profile. Note that benchmarks may include profile upgrade routines.
- power_user -- a pre-recorded profile that represents a power user's profile. Note that benchmarks may include profile upgrade routines.
A path to a profile to use. This is useful with generated profiles.
Because the canned profile-types may exhibit unrealistic behavior in benchmarks (due to profile initialization or update routines), it is useful to be able to generate a deterministic profile for benchmarking. To do so:
$ tools/perf/generate_profile --profile-type-to-generate=<PROFILE_TYPE> --output-dir=/path/to/output/profile
may be small_profile, theme_profile or many_extensions_profile.
Generated profile output directories should then be passed to --profile-dir. For example, to run the session_restore benchmark:
$ tools/perf/generate_profile --profile-type-to-generate=small_profile --output-dir=out/Release/generated_profile
$ tools/perf/run_benchmark --profile-dir=out/Release/generated_profile/small_profile session_restore.cold.typical_25