Setup

NOTE: The README assumes an environment where the code is cloned, edited, and built locally, and is deployed on a remote cloudlab cluster. Adjust the commands as necessary depending on the actual execution deployment environment.

Cloning Repositories

Clone the following repositories. Refer to this section below for a description on different cloning options

git clone git@github.com:gt-epl/feo.git
git clone git@github.com:gt-epl/feo-loadgen.git
git clone git@github.com:gt-epl/feo-notebooks.git

Modifying Configurations

Append the following entry to a sshconfig file (e.g. ~/.ssh/config) The following is an example configuration. Note the alias clabcl0 defined next to Host. This alias is an important key in loadgen/run_load.py. For the scripts to work correctly, the last host should be clabsvr. You may adjust the scripts if you wish to use a different nameset.

Host clabcl0
  User {your cloudlab username}
  Hostname {your cloudlab hostname}
  Port {your cloudlab portnumber}
  IdentityFile {path to the private SSH key registered in cloudlab}
  StrictHostKeyChecking=no
  UserKnownHostsFile ~/.ssh/clab_hosts
  ServerAliveInterval 120

Host clabcl1
...

Host clabsvr
...

On the loadgen/run_load.py file, adjust the configurations in the commented area.

On feo/config.template.yml, adjust the peer address & controller address so that the nodes can send messages to each other. Make sure that the order of the peer address matches the order of the nodes in 'hosts'.

On each profile file under loadgen/profiles, the hostname in the first column should match the names in the list hosts defined in run_load.py.

Copying files to each node

There are 3 options to run the files

1. Run loadgen/run_load.py (and eventually) ./utils/sync.sh locally. In this case you do not have to copy an sshconfig file or a sshkey file to a remote machine. loaden/run_load.py takes care of copying any binary and/or application, util files.

Note: In run_load.py, CONFIG_EXEC_LOCAL must be set to True. Also, the local node must have go & python installed to build/run the scripts.

2. Copy this directory (feo) into clabsvr node. We run loadgen/run_load.py locally, but the script will run utils/sync.sh remotely.

Note: This requires us to copy a SSH private key to clabsvr, so that the clabsvr node can ssh into other nodes. Create a dedicated, NON_CRITICAL SSH key for this instance!

cd ../
rsync {path to the sshconfig file defined above} clabsvr:~/.ssh/{path to sshconfig file}
rsync {path to the ssh identity file (privatekey) defined above} clabsvr:~/.ssh/{path to identity file}
rsync -avz ./feo clabsvr:~/  OR  ssh clabsvr; cd ~; git clone git@github.gatech.edu:faasedge/feo.git

3. Copy both ../feo and ../loadgen to clabsvr node. This allows us to avoid the need to setup a build environment locally.

Note: In run_load.py, CONFIG_EXEC_LOCAL must be set to True. Also, the results residing in clabsvr:RESULT_DIR must be moved to a persistent storage.

In addition to the above command, run:

rsync -avz ./loadgen clabsvr:~/  OR  ssh clabsvr; cd ~; git clone git@github.gatech.edu:faasedge/loadgen.git

Run evaluations

python run_load.py profile.csv

If ACTUAL and FETCH_RESULTS are set to True in run_load.py, the results will appear in RESDIR, also defined in run_load.py.

Use the notebooks in feo-notebooks to plot the results.

How to check if deployment is correct?

On each node,

• To check if openwhisk has been deployed

docker ps

• To check if an action has been deployed

bash list.sh http://localhost:3233

• To check if an action is correctly invoked

bash apps/copy/invoke_action.sh http://localhost:3233

• To check if the inter-node latency has successfully been set using set_latency.sh, simply use ping