The code in this repository can be used to reproduce the experimental evaluations seen in the paper
Johan Ruuskanen, Anton Cervin, "Distributed online extraction of a fluid model for microservice applications using local tracing data", to appear at IEEE CLOUD 2022
The experiments are run on the FedApp sandbox, while the models are extracted and evaluated using julia-1.7.0.
In order to recreate the figures seen in the paper, you must first deploy the sandbox with the example application, run the experiments and finally extract the models.
Access to an OpenStack cloud is required to deploy the sandbox out-of-the-box. The sandbox can possibly also be deployed on other infrastructures, but this requires some tinkering.
Follow the steps in shown in the sanbox repo, and deploy the gateway with 4vCPU and 16 Gb of RAM, along with 2 cluster each with 4 virtual machines, each with 4vCPU and8 Gb of RAM.
Finally, add a delay of 25 ms between the two clusters using TC netem.
To deploy the example application on the FedApp sandbox, begin by copying the application/facedetect_3s folder to the gateway.
We used the UMass face detection database to provide the necessary images for loading the application. Download it and extract it to the application/data folder on the gateway.
To avoid time-dependencies of requests fetching images not yet stored, we pre-populated the storage service with the possible images fetched. To do this, simply run application/facedetect_3s/apps/storage/populate.py before proceeding.
In application/facedetect_3s/apps, change the gitlab repository in the build_and_push.sh script to point to a gitlab container registry that you can access. Then run the following script to build and push all necessary container images.
chmod +x build_and_push.sh
./build_and_push.sh
Also, change the gitlab repository in each of the 4 service YAMLs
backend-v1.yaml
backend-v2.yaml
frontend.yaml
storage.yaml
The application can then be deploy with
python3 deploy.py
To later remove it, simply run
python3 remove.py
The application can be reached on port 3001 on the gateway, and to access it simply perform a port-forward using e.g. ssh ssh -L 3001:localhost:3001 ubuntu@GATEWAY_IP and then visit http://localhost:3001 on your local computer.
After the application has been deployed, visit the application/facedetect_3s/load_generation folder. In run_experiment.py, choose which experiment to run, and then run it with
python3 run_experiment.py
The generated logs in application/facedetect_3s/load_generation/logs contains the necessary tracing data for the model extraction.
After running the workload and scaling experiments, copy the gathered logs/ folders to the model_extracion folder on your local computer.
The model extraction and prediction experiments can be performed in the model_extraction folder. They are implemented in Julia and tested with Julia-1.7.0. For the plotting, it further requires Python with matplotlib.
To extract a model and perform predictions, start Julia in the model_extraction folder and type
] activate .
] instantiateto setup the environment.
The workload experiments are found in examine_model_diffLoad.jl, while the scaling experiments are found in examine_model_scaling.jl. Simply change the datafolder = "" in the two experiment scripts to point to the correct tracing log folder from the experiments and run the script.