TrainTicket Utility Artifacts and Documentation

This repository contains the artifacts created during a student research project at the University of Stuttgart. They contain helper scripts and configuration files for using and loadtesting the TrainTicket benchmark system, partially adapted/ported from work provided by Martin Straesser at the University of Würzburg. TrainTicket was developed at the Fudan University and can be found on GitHub.

Additionally, we provide some documentation based on our findings when running TrainTicket in our study.

Contents

• deployments/ contains custom deployment files we created to run TrainTicket.
• loadgenerator/docker-deployment contains a docker-compose and dockerfiles to run the load generator. The Steady State and Overload Profiles were created with LIMBO.
• loadgenerator/load-profiles contains the load profiles used for load testing TrainTicket.
• loadgenerator/user-profiles contains two user profiles defining requests to TrainTicket. The Steady State Profile is a set of requests simulating a user's interaction with TrainTicket. The Overload Profile sends GET, POST and DELETE requests to the food service.
• shell-scripts/ contains scripts to clean the databases of TrainTicket between load testing runs.
• traces/ contains a single trace for the /preserve endpoint in TrainTicket, gathered with Jaeger.

Additional Documentation

Launching TrainTicket (docker-compose, Linux)

Start TrainTicket w/docker-compose:

sudo docker-compose -f deployment/docker-compose-manifests/docker-compose-with-jaeger.yml up

Shutting down & cleaning up after TrainTicket - shuts down containers and removes unused networks and volumes.

sudo docker-compose -f deployment/docker-compose-manifests/docker-compose-with-jaeger.yml down
sudo docker network prune -f
sudo docker volume prune -f

Launching TrainTicket on Kubernetes with Minikube

Starting Minikube with more resources

minikube start --memory 16384 --cpus 6

Deleting and restarting the Minikube cluster (for a clean deploy)

minikube stop
minikube delete
minikube start --memory 16384 --cpus 6

Running the Minikube dashboard (helps with monitoring pods)

minikube dashboard

Deploying TrainTicket

cd deployment/kubernetes-manifests/quickstart-k8s
kubectl apply -f quickstart-ts-deployment-part1.yml
kubectl apply -f quickstart-ts-deployment-part2.yml
kubectl apply -f quickstart-ts-deployment-part3.yml

To improve the performance of TrainTicket, try the deployment without resource limits for services found in deployments/k8s-no-resource-limits/.

Loadgenerator usage

First run load generator with:

java -jar httploadgenerator.jar loadgenerator

Then run loadgenerator director with:

java -jar httploadgenerator.jar director --load tinyRequests.csv -o testlog.csv --yaml scenario01-user-profile.yaml

Further Reading

Fault Analysis and Debugging of Microservice Systems: Industrial Survey, Benchmark System, and Empirical Study - Xiang Zhou, Xin Peng, Tao Xie, Jun Sun, Chao Ji, Wenhai Li, Dan Ding.

Run-Time Prediction of Power Consumption for Component Deployments - Jóakim von Kistowski, Maximilian Deffner, Samuel Kounev.

Modeling and Extracting Load Intensity Profiles - Jóakim Von Kistowski, Nikolas Herbst, Samuel Kounev, Henning Groenda, Christian Stier, Sebastian Lehrig