Experiment runner framework for Dispersy and Tribler.

After setting up the workspace (see below), simply call gumby/run.py passing your experiment's config file path as argument.

Example:

gumby/run.py gumby/experiments/dummy/local_prun.conf

To create a new experiment config, just run this:

gumby/scripts/generate_config_file.py gumby/experiments/my_experiment/new_experiment.conf

This will parse all of Gumby resources and create a self-documented config skeleton for you. Open the file and read the instructions there.

• Clear output dir
• Sync workspace dir with remote nodes
• Run local and remote setup scripts concurrently (see local_setup_cmd and remote_setup_cmd config options in your config file)
  • If any of them fail, the experiment will be aborted.
• Start the tracker in the background (optional, see tracker_cmd).
  • Spawn the tracker and keep it running during the whole experiment run time.
  • If the tracker dies during the experiment execution, the experiment will be aborted.
  • When the experiment finishes, the tracker will be killed automatically.
• Start the experiment server in the background (optional, see experiment_server_cmd).
  • If the experiment server exits (default behavior after giving the go signal to all the instances) the experiment will not be aborted.
• Start both local and remote process instances in parallel (see local_instance_cmd and remote_instance_cmd).
• Wait for all the instances to die.
• Collect all the data from the remote output dirs.
• Run the post-process script locally to generate graphs and whatnot (optional, see post_process_cmd)

Remember that the output dir will be wiped out at every experiment execution. If you want to keep the output of several runs and you aren't using Jenkins or similar, you could run your experiments with something like GUMBY_OUTPUT_DIR=$PWD/output_$(date "+%y-%m-%d_%H:%M:%S") ./gumby/run.py gumby/experiments/.......

Gumby expects the following directory tree:

workspace/: The workspace dir contains everything that the experiment will need during its execution (including Gumby), it can be named as you like.

workspace/gumby/: Here is where you should clone the gumby repo, it has to be located in the root of the workspace with this specific name.

workspace/tribler/: If your experiments use code from Tribler, clone the repository in this location so all the helper scripts can find it.

workspace/dispersy/: Idem.

workspace/output/: Experiment output location, this directory will be cleared/created automatically whenever the experiment is started, so take care of copying anything you want to keep before restarting an experiment.

A typical set up/execution session up would look similar to this:

ssh das4
cd /var/scratch/$USER
mkdir my_workspace
cd my_workspace
git clone https://github.com/Tribler/gumby
git clone https://github.com/Tribler/tribler
gumby/run.py gumby/experiments/dispersy/allchannel.conf
ls -R output/

Experiment entry point, must receive an experiment config file as only argument.

It contains all the settings needed to run an experiment using this framework.

It will usually be stored into experiments/ExperimentName/experiment.conf If you want to have several variations of the same experiment, store several config files in the experiment dir.

(All relative to Gumby's repository root )

Experiments go in there, in a folder named after each experiment or experiment. If you want several variations of the same experiment, just create a .conf file for each of them. IE:

• experiments/my_crazy_experiment/crazy.conf
• experiments/my_crazy_experiment/crazier.conf
• experiments/my_crazy_experiment/craziest.conf

If your experiment needs to use a custom script, just store it in this directory, it will be automatically added to the PATH. (the same goes for PYTHONPATH)

Experiments waiting to be ported to Gumby.

Scripts that can be used as components to build experiments. If you create a script generic enough to be useful to others, please send a pull request to add it to the collection.

Please, check the documentation in the config file template.