Replication Package – Topology and Scale Effects on Performance and Energy in μBench Microservices

This repository supports replication of the study on topology and scale effects on performance and energy in μBench microservices. The layout is aligned with the paper structure:

• Section 3: Experiment Planning — Topology definitions, workmodels, and service-graph choices. These are not in a separate directory; they are documented below and live inside 4_experiment_execution/.
• Section 4: Experiment Execution — 4_experiment_execution/
• Section 5: Results and Analysis — 5_results_data/ (data) and 5_results_analysis/ (notebooks and outputs)

---

How to cite the dataset

If this replication package is helping your research, consider citing it as follows:

@misc{ReplicationPackage,
    title = {{An Empirical Study on How Architectural Topology Affects Microservice Performance and Energy Usage}},
    author = {{Irena Ristova, Vincenzo Stoico}},
    year = {2026},
    publisher = {GreenArch 2026: The First International Workshop on Software Architecture for Green Sustainable Carbon-aware Software Systems},
    url = {https://github.com/S2-group/greenarch26-topology-impact},
    note = {Accessed: 2026-03-27}
}

(Update the note field with your date of access when citing.)

---

Section 3: Experiment Planning

The material that corresponds to Section 3 of the paper (experiment planning: topologies, workmodels, service graphs) is contained in 4_experiment_execution/:

• Workmodel definitions: 4_experiment_execution/muBench/Examples/ — The 18 workmodel JSON files used in the experiment (e.g. workmodel-serial-{5,10,20}services.json, workmodel-parallel-*, workmodelA*, workmodelC*, workmodelC-multi*, workmodelD*). These define topology patterns and system sizes.
• Topology–workmodel mapping: 4_experiment_execution/experiment-runner/examples/mubench-benchmarking/RunnerConfig.py — The function get_workmodel_path() maps each (topology, size) to a workmodel file. The factors (6 topologies × 3 sizes) and design are documented in that folder’s README.md and EXPERIMENT_ARCHITECTURE.md.
• Deployment configuration: 4_experiment_execution/muBench/Configs/ — Configs used by the deployer (e.g. K8sParameters.json).

There is no 3_experiment_planning/ directory; the planning artifacts above are used directly by the execution in Section 4.

---

Section 4: Experiment Execution — 4_experiment_execution/

Paper section: Section 4 (Experiment Execution).

This directory contains everything needed to re-run the experiment: Experiment Runner, a trimmed copy of muBench, and the experiment definition (RunnerConfig.py). It includes the workmodels and scripts used for deployment, load generation (Locust), and measurement (Prometheus, EnergiBridge).

• Contents: experiment-runner/, muBench/, and the mubench-benchmarking example (RunnerConfig.py, EXPERIMENT_ARCHITECTURE.md).
• Documentation: 4_experiment_execution/README.md, 4_experiment_execution/HOW_TO_RUN_THE_EXPERIMENT.md.

---

Section 5: Results and Analysis

5_results_data/ — Result data (Section 5)

Paper section: Section 5 (Results and Analysis — data).

Contains the experiment outputs: run_table.csv (aggregated metrics per run), raw_runs/ (per-run Locust, Prometheus, and EnergiBridge outputs), and RUN_TABLE_COLUMNS_EXPLANATION.md. One raw EnergiBridge file (run_15_repetition_5/energibridge_output.csv) exceeds GitHub’s 100 MB limit and is omitted from the repo; its aggregated values are in run_table.csv, and the raw file can be reproduced by re-running that configuration (see 4_experiment_execution/HOW_TO_RUN_THE_EXPERIMENT.md).

• Documentation: 5_results_data/README.md.

5_results_analysis/ — Analysis notebooks and outputs (Section 5)

Paper section: Section 5 (Results and Analysis — figures and tables).

Contains Jupyter notebooks that produce the paper’s figures and tables from 5_results_data/run_table.csv. Outputs are written to 5_results_analysis/figures/ and 5_results_analysis/tables/.

• Documentation: 5_results_analysis/README.md.

---

Replicating the data analysis (Section 5 only)

To reproduce the paper’s tables and figures from the provided data without re-running the experiment, run the Jupyter notebooks in 5_results_analysis/ (Boxplots, Descriptive_Statistics, Structural_Groups_Analysis). Input: 5_results_data/run_table.csv. Outputs: 5_results_analysis/figures/, 5_results_analysis/tables/.

See 5_results_analysis/README.md for how to activate the virtualenv and start Jupyter.

---

Running the entire replication (Sections 4 and 5)

1. Section 4 — Experiment execution: Follow 4_experiment_execution/HOW_TO_RUN_THE_EXPERIMENT.md to install prerequisites, set up the server, start tunnels, and run the Experiment Runner. Results are produced under the runner’s output path; the package’s 5_results_data/ is a snapshot of that output.
2. Section 5 — Analysis: Use the run table (from step 1 or the provided 5_results_data/run_table.csv) and run the analysis notebooks as described in 5_results_analysis/README.md.

---

About

Replication package for the study on topology and scale effects on performance and energy in μBench microservices (Experiment Runner, muBench, Locust, Prometheus, EnergiBridge).