Artifact for ICSA submission "Let LLMs Autoscale Microservices: A Quantitative Evaluation of Architectures"

The dataset contains evaluation experiments for different architectural alternatives employing large language models in different roles.The dataset contains specifically four parts: Scaling architectures and deployment scripts of the tested alternatives under /scaling-architectures-deployment_and_runtime_codeThe raw results collected which also are discussed in the paper under /resultsKubernetes Deployment definitions of TeaStore and Locust (the load generator) under /k8s_deploymentLocust container with all the different load pattern definition under /docker_container_locustThe dataset has been collected by Tim Summerer in the following infrastructure/architecture: - Hyperscaler: Google Cloud (Google Cloud Research Credits program with the award GCP19980904)- Type of Deployment: Google Kubernetes Engine (GKE) Autopilot- Kubernetes version 1.32.x (1.32.4-gke.1415000 to 1.32.6-gke.1125000)- Locust version: 2.37.11- Python versions: 3.12.11 for vllm v0.10.0 on Fedora 42, 3.13.3 for scripts and 3.13.5 for Jupyter Notebooks- Teastore Docker container version: 1.4.2- Istio (service mesh) version: 1.26.2 (includes Prometheus 27.8.0)- KEDA version: v2.17.2