Enoki: Stateful Distributed FaaS from Edge to Cloud

Enoki is a prototype FaaS platform for the edge-cloud that integrates replicated data access.

Research

If you use this software in a publication, please cite it as:

Text

T. Pfandzelter and D. Bermbach, Enoki: Stateful Distributed FaaS from Edge to Cloud, Proceedings of the 2nd International Workshop on Middleware for the Edge (MiddleWEdge '23), Bologna, Italy. ACM New York, NY, USA, 2023.

BibTeX

@inproceedings{pfandzelter2023enoki,
    author = "Pfandzelter, Tobias and Bermbach, David",
    title = "Enoki: Stateful Distributed FaaS from Edge to Cloud",
    booktitle = "Proceedings of the 2nd International Workshop on Middleware for the Edge (MiddleWEdge '23)",
    month = dec,
    year = 2023,
    publisher = "ACM",
    address = "New York, NY, USA",
    location = "Bologna, Italy",
    doi = "10.1145/3630180.3631203"
}

A preprint is available on arXiv. For a full list of publications, please see our website.

License

The code in this repository is licensed under the terms of the MIT license.

Usage

To run Enoki locally, we recommend checking out the prep-simple.sh and run-simple.sh scripts. At its core, Enoki requires a running fred instance and certificates to start.

Experiments

To replicate the experiments used in our paper, use the provided scripts.

Prerequisites

To run experiments, you will need:

• a local machine running Debian 11 or a similar Linux distribution (modifications are needed for other distributions and macOS)
• an active Google Cloud Platform project (note the project ID)

Install these dependencies on your local machine:

sudo apt-get update
sudo apt-get install apt-transport-https ca-certificates gnupg curl sudo rsync jq zip bc multitail python3-pip
python3 -m pip install -r requirements.txt

Install the gcloud CLI according to the documentation. Configure the gcloud CLI for your project with gcloud auth login --no-launch-browser.

Install terraform (>v1.0) as outlined in the documentation. Configure your Google Cloud credentials for Terraform with gcloud auth application-default login and then run terraform init.

Install Go (>=v1.20):

wget <https://go.dev/dl/go1.20.7.linux-amd64.tar.gz>
rm -rf /usr/local/go && sudo tar -C /usr/local -xzf go1.20.7.linux-amd64.tar.gz
echo "export PATH=$PATH:/usr/local/go/bin" >> .bashrc
source .bashrc
go version

Running Experiments

Before you run any experiments, make sure to update the settings in variables.tf. Configure your Google Cloud project ID and your preferred Google Cloud Compute region and availability zone.

For each of the four experiments outlined in our paper (simple, scale, replication, and befaas), we include an automated experiment-*.py Python script. The scripts provide only limited output on stdout, with most output piped to the output directory.

Run each script in a separate screen and then monitor output using multitail:

# open a new screen
screen
# start the experiment
./experiment-NAME.py
# exit the screen with CTRL-A + D

# monitor progress
multitail -cT ANSI -iw "output-NAME/*" 5

Resources on Google Cloud platform are automatically started and destroyed for you. Note that it might be necessary to remove old results in the results directory before you can run new experiments.

Analysis

To run the analysis and generate graphs, use the included graphs-*.py scripts. Note that for the BeFaaS experiments, a separate cleaning step is necessary:

./sortlogs-befaas.py ./results/results-befaas/raw ./results/results-befaas