This repository contains the source code for the RUNE hybrid FaaS runtime described in the Middleware'25 paper A Hybrid Runtime for Function-as-a-Service at the Edge.
RUNE is implemented as FaaSEdge Core (fecore), which is the core runtime for the FaaSEdge project from the Embedded Pervasive Lab at Georgia Tech. This document focuses on getting RUNE up-and-running for the purpose of evaluation.
The main README for the fecore/RUNE project can be found in README_fecore.md.
The instructions and setup described in this document have been tested on Cloudlab c220g1 machines. The fastest way to evaluate RUNE is to use our pre-built Cloudlab image:
- You will need to register for an account at Cloudlab
- Once you have registered, visit the profile page for our RUNE-ae Profile
- Select the
Instantiatebutton - You can evaluate RUNE using a single node
- Select a project to attach this experiment to (if you do not already have a project to use, one can be created by selecting your username in the top-right corner and then "Start/Join Project")
- Select "Next" and then "Finish" to launch a new experiment
Cloudlab will provision a bare metal c220g1 machine and load our pre-built Linux image that contains most dependencies needed to run RUNE.
After the machine has been brought online, SSH into it and change directory to /project/RUNE. You should find the files needed to complete the setup in this directory. NOTE: If you do not see any files in this directory, or if you are not using our Cloudlab image, you will need to manually download the latest release of RUNE and run the SETUP.sh script found within.
Run the SETUP.sh script to finish installation of remaining of dependencies and configs.
Note that you can also set up an environment manually by following instructions in README_fecore.md and SETUP.md. The provided SETUP.sh script can also be readily adapted to other Ubuntu 20.04 LTS installations.
You will need at least two console sessions to evaluate RUNE. We recommend using a program like tmux or screen to split the console into two halves so that you can run commands and view output from RUNE at the same time.
In the first console session, start the RUNE server with the following commands:
sudo su -cd /usr/local/binsecret_mount_path=/var/lib/fecore/secrets fecore provider
In the second console session, deploy example functions with the following commands:
cd /mnt/faasedge/bin/
faas-cli -g 10.62.0.1:8081 deploy --image achgt/thumbnailer-n:latest --name thumbnailer-n --label ctrType=native
faas-cli -g 10.62.0.1:8081 deploy --image=thumbnailer-w --name thumbnailer-w --label ctrType=wasm
faas-cli -g 10.62.0.1:8081 deploy --image hybrid --name thumbnailer-h --label ctrType=hybrid --label sandboxes=thumbnailer-n,thumbnailer-w
faas-cli -g 10.62.0.1:8081 deploy --image achgt/encrypt-p:latest --name encrypt-n --label ctrType=native
faas-cli -g 10.62.0.1:8081 deploy --image=encrypt-w --name encrypt-w --label ctrType=wasm
faas-cli -g 10.62.0.1:8081 deploy --image hybrid --name encrypt-h --label ctrType=hybrid --label sandboxes=encrypt-n,encrypt-w
faas-cli -g 10.62.0.1:8081 deploy --image achgt/compression-n:latest --name compression-n --label ctrType=native
faas-cli -g 10.62.0.1:8081 deploy --image=compression-w --name compression-w --label ctrType=wasm
faas-cli -g 10.62.0.1:8081 deploy --image hybrid --name compression-h --label ctrType=hybrid --label sandboxes=compression-n,compression-w
The above commands will deploy Native, WASM, and Hybrid versions of three example functions described in the RUNE paper (Thumbnailer, Encrypt, and Compression). The Native versions are available as pre-built images via DockerHub and the pre-built WASM versions are included in the latest release of RUNE under the examples directory.
You can test RUNE's hybrid container capability by invoking a hybrid function deployment via curl, e.g.: curl -vk http://10.62.0.1:8081/function/compression-h
The example functions have a default policy that specifies using a WebAssembly (wasm) container for cold starts and a native container for warm starts. The first time you invoke a hybrid function with curl you should see output similar to the following:
* Trying 10.62.0.1:8081...
* Connected to 10.62.0.1 (10.62.0.1) port 8081 (#0)
> GET /function/compression-h HTTP/1.1
> Host: 10.62.0.1:8081
> User-Agent: curl/7.74.0
> Accept: */*
>
* Mark bundle as not supporting multiuse
< HTTP/1.1 200 OK
< Container-Name: compression-w_ebfaef91-6224-438a-901a-9c2d3b10e9f8_w
< Container-Type: hybrid
< Content-Type: text/plain
< Invocation-Elapsed: 272
< Misc-Stats: 27.773000,18.215000,373.543000,108.675000
< Request-Id: compression-h_bd842ef8-8ea1-4118-a96c-2898d6b139b3
< Runw-Setup: 1
< Server: runw
< Setup-Time: 23
< Startup-Type: cold
< Date: Sat, 18 Oct 2025 06:48:01 GMT
< Content-Length: 9
<
Success
* Connection #0 to host 10.62.0.1 left intact
Two important fields to note in the output are Container-Name and Startup-Type. Here we see the container name starts with compression-w, indicating a WebAssembly container was used. The Startup-Type is cold.
If you run the curl command again, you should see output similar to the following:
* Trying 10.62.0.1:8081...
* Connected to 10.62.0.1 (10.62.0.1) port 8081 (#0)
> GET /function/compression-h HTTP/1.1
> Host: 10.62.0.1:8081
> User-Agent: curl/7.74.0
> Accept: */*
>
* Mark bundle as not supporting multiuse
< HTTP/1.1 200 OK
< Container-Name: compression-n_b6015450-6462-44aa-aa41-c92bec5b7ce8_n
< Container-Type: hybrid
< Content-Length: 8
< Content-Type: text/plain; charset=utf-8
< Date: Sat, 18 Oct 2025 06:48:03 GMT
< Invocation-Elapsed: 157
< Request-Id: compression-h_47d43fad-c660-4e2d-8edb-cc3f1d878520
< Setup-Time: 0
< Startup-Type: warm
<
1227561
* Connection #0 to host 10.62.0.1 left intact
In this second output we see the Container-Name now begins with compression-n, indicating a native container, and the Startup-Type is now warm.
The warm native container will be destroyed after 10 seconds of inactivity, necessitating another cold start. The server output provides detailed information such as how each container was spawned, how much time was required for setup and execution, and when the idle container was deleted.
When you are finished evaluating RUNE, pressing Ctrl+c in the console session running fecore should clean up any running containers and perform a graceful shutdown.
Research involved in the creation of this project was published in ACM Middleware 2025. A PDF copy of the paper can be found here.
@inproceedings{10.1145/3721462.3770785,
author = {Hall, Adam and Ramachandran, Umakishore},
title = {A Hybrid Runtime for Function-as-a-Service at the Edge},
year = {2025},
isbn = {9798400715549},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3721462.3770785},
doi = {10.1145/3721462.3770785},
abstract = {Contemporary Function-as-a-Service (FaaS) platforms leverage lightweight container-based virtualization to host instances of applications on-demand. This on-demand model is a good fit for computing environments where resource efficiency is desirable (such as the Cloud) or necessary (such as the Edge). However, the use of containers introduces significant delays to end-to-end service times in the form of cold start latency. As a workaround, FaaS platforms employ techniques like pre-warming containers or keeping idle containers alive to reduce the number of cold starts required. These techniques are effective, but greatly reduce platform resource efficiency, in turn making them less suitable for target computing deployments. Alternatives to containers like WebAssembly offer significantly lower cold start time, but have slower execution speeds than native code and cannot serve as a replacement to containers. We argue that combining the strengths of both containers and WebAssembly into a single hybrid runtime for FaaS platforms can improve both the end-to-end response time latencies and resource efficiency. We demonstrate this principle through proof-by-construction and introduce RUNE: a hybrid RUNtime for FaaS at the Edge. We evaluate RUNE empirically and show that its hybrid approach reduces tail latency response times at the 99th percentile by more than half while also improving memory utilization of the underlying platform by reducing the number of idle containers.},
booktitle = {Proceedings of the 26th International Middleware Conference},
pages = {469–481},
numpages = {13},
keywords = {serverless, FaaS, edge computing},
location = {Vanderbilt University, Nashville, TN, USA},
series = {Middleware '25}
}