Description

Hello, I am an electric benchy, I can't go very fast or very far and if you drive me people will think you are awesome.

Benchmark Definition

Each benchmark defines basically:

In which images it need to be run
With which vms
and what code does the vm needs to evaluate to run the benchmark

One benchmark can be run on several images with several vms. Benchy will take care of that, you only need to specify the configuration.

Quick Start

I am a framework for doing benchmarks in Pharo.

Running all benchmarks

Just run the runAll.sh script. That will:

Download all the pharo images that are defined in the /images folder
Then it will built the vms defined in the /vms folder
then it will run the benchmarks defined in the benchs folder

Running options

Running run.sh will list options to run benchmarks.

To list the available benchmarks ./run.sh list

To run specific benchmarks in all images and all VMs

./run.sh bench slopstone
./run.sh bench richards

To plot existing results ./run.sh bench plot

Output

Benchy will create a csv file with the time of the execution and two log files with the logs that happened during the execution. In your Pharo code you can to anObj trace to leave traces.

Running your own benchmarks

Parameters

It is possible to specify which directory containing benchmarks you want benchy to use. To do so, you must write BENCHES_SCRIPT_DIR=<path to your directory> ./runAll.sh. The path can be either relative or absolute. If not specified, BENCHES_SCRIPT_DIR is by default the benchs folder of benchy.

By default, benchy runs 30 iterations of each benchmarks. It is possible to customize this by using : ITERATIONS=<number of your choice> ./runAll.sh.
For instance with ITERATIONS=1 ./runAll.sh, each benchmark will run once.

You can also change the number of iterations locally. If you run ITERATIONS=1 BENCHES_SCRIPT_DIR=myDir ./runAll.sh with myDir being a directory containing several benchmarks, including the following one, it will run this benchmark file with 2 iterations but all the others with 1 iteration (look further down to see how to create benchmark configuration).

#!/bin/bash

# Configuration
IMAGES="Pharo11"
VMs="latest9"

# Command
PHARO_CMD="eval 1+1"

# Iterations
ITERATIONS=2

# Run
runBenchs

Note that if your number is 0 or less, benchy won't produce anything.

You can also add VM options when running runAll.sh with VM_PARAMETERS=<parameter> ./runAll.sh.
Those options are available with <vm> --help. Like ITERATIONS, you can specify them locally.

Executing your benchmars

The benchmarks that are here by default are made to test the performance of Pharo. If you want to run your own benchmarks you need to:

Add a .sh file in the /benchs folder with the configuration of your benchmark.
[Optional] Add a .sh file in the /vms folder with the instructions to download your vm. If not you can use one of the already present vms.
Execute runBenchs.sh name_of_your_benchmark with the name.s of your benchmark.s

Create your own configuration

You can create your own benchmark configuration by creating a .sh file using the following template:

#!/bin/bash

# Configuration
IMAGES="<the images you want your benchmark to run in>"
VMs="<the VMs you want your benchmark to run with>"

# Command
PHARO_CMD="<the command you want the VM to execute>"

# Run
runBenchs

and then add it to the /benchs folder.

You can also chain the following lines to run the same command with other images and/or VMs:

#!/bin/bash

# Command
PHARO_CMD="<the command you want the VM to execute>"

# First configuration
IMAGES="<the images you want your benchmark to run in>"
VMs="<the VMs you want your benchmark to run with>"

# Run
runBenchs


# Second configuration
IMAGES="<other images>"
VMs="<other VMs>"

# Iterations
ITERATIONS=3

# Run
runBenchs

Be sure not to forget to rewrite runBenchs so that the benchmark reruns with the new images and/or VMs.

Dependencies

In order to run benchParamTest.sh, you need to download shunit2 and add the path to the shunit2 file to your PATH.

Pharo Installation

EpMonitor disableDuring: [ 
	Metacello new
		baseline: 'BenchHistory';
		repository: 'github://tesonep/benchy:main';
		load ]

More info

In the /src folder there is code to make plots for the benchmarks.

tesonep/benchy

Table of Contents