EXPERIMENTS.md

Experiments

Each file in the experiment directory runs Ansible code:

• ansible.cfg: tells Ansible where the roles directory is and specifies how to use SSH.

• cleanup.yml: cleans up after the experiment (e.g., removing containers, deleting directories, etc.)

• experiment.yml: bash-like syntax for running the benchmark.

• hosts.template: specifies the layout of the cluster, like which services are on which nodes.

• site: tells Ansible where to look for roles. For every system in your experiment (e.g., Hadoop, Swift, etc.), you must have a site.yml file.

To add a new experiment, you must add a role and an experiment configuration.

Adding Experiments

1. Create a new Git repository for your experiments:

2. Copy the templates and sites for the systems you want to use:

   cp -r srl-roles/hadoop/test <my_benchmark>/experiment1
   cp srl-roles/swift/test/site/site.yml <my_benchmark>/experiment1/site/swift.yml

3. Edit your experiment and setup your group variables:

   cd <my_benchmark>/experiment1
   vim experiment.yml
   vim group_vars/all

Adding Roles

A role is code that installs/deploys a system. To see an example, see the Swift role. If you look in a role directory, you will see two types of directories:

1. dockerfiles: Dockerfile code for building the image. Docker images have all the packages and environment variables needed to run the system.

2. all other directories: Ansible scripts (*.yml) that either deploy or build the system. These are used in the site.yml files you setup above.

EOF