Toolkit to run Python benchmarks
Clone or download
Latest commit 7cdcd58 Jun 3, 2018



Latest release on the Python Cheeseshop (PyPI) Build status of perf on Travis CI

The Python perf module is a toolkit to write, run and analyze benchmarks.


  • Simple API to run reliable benchmarks
  • Automatically calibrate a benchmark for a time budget.
  • Spawn multiple worker processes.
  • Compute the mean and standard deviation.
  • Detect if a benchmark result seems unstable.
  • JSON format to store benchmark results.
  • Support multiple units: seconds, bytes and integer.


To run a benchmark use the perf timeit command (result written into bench.json):

$ python3 -m perf timeit '[1,2]*1000' -o bench.json
Mean +- std dev: 4.22 us +- 0.08 us

Or write a benchmark script

#!/usr/bin/env python3
import perf

runner = perf.Runner()
runner.timeit(name="sort a sorted list",
              stmt="sorted(s, key=f)",
              setup="f = lambda x: x; s = list(range(1000))")

See the API docs for full details on the timeit function and the Runner class. To run the script and dump the results into a file named bench.json:

$ python3 -o bench.json

To analyze benchmark results use the perf stats command:

$ python3 -m perf stats bench.json
Total duration: 29.2 sec
Start date: 2016-10-21 03:14:19
End date: 2016-10-21 03:14:53
Raw value minimum: 177 ms
Raw value maximum: 183 ms

Number of calibration run: 1
Number of run with values: 40
Total number of run: 41

Number of warmup per run: 1
Number of value per run: 3
Loop iterations per value: 8
Total number of values: 120

Minimum:         22.1 ms
Median +- MAD:   22.5 ms +- 0.1 ms
Mean +- std dev: 22.5 ms +- 0.2 ms
Maximum:         22.9 ms

  0th percentile: 22.1 ms (-2% of the mean) -- minimum
  5th percentile: 22.3 ms (-1% of the mean)
 25th percentile: 22.4 ms (-1% of the mean) -- Q1
 50th percentile: 22.5 ms (-0% of the mean) -- median
 75th percentile: 22.7 ms (+1% of the mean) -- Q3
 95th percentile: 22.9 ms (+2% of the mean)
100th percentile: 22.9 ms (+2% of the mean) -- maximum

Number of outlier (out of 22.0 ms..23.0 ms): 0

There's also:

  • perf compare_to command tests if a difference is significant. It supports comparison between multiple benchmark suites (made of multiple benchmarks)

    $ python3 -m perf compare_to py2.json py3.json --table
    | Benchmark | py2     | py3                          |
    | timeit    | 4.70 us | 4.22 us: 1.11x faster (-10%) |
  • perf system tune command to tune your system to run stable benchmarks.

  • Automatically collect metadata on the computer and the benchmark: use the perf metadata command to display them, or the perf collect_metadata command to manually collect them.

  • --track-memory and --tracemalloc options to track the memory usage of a benchmark.

Quick Links

Command to install perf on Python 3:

python3 -m pip install perf

perf supports Python 2.7 and Python 3. It is distributed under the MIT license.