Replication package for "Leveraging Network Methods for Hub-like Microservice Detection"

This is the replication package for our paper "Leveraging Network Methods for Hub-like Microservice Detection".

Contents

This repository contains the following files:

• Projects List of analyzed OSS microservice projects take from Bakhtin et al. [1]
• Raw Raw SDG networks taken from Bakhtin et al. [1]
• Metrics Network metrics computed from raw data
  • CentralityMetrics.csv All the computed network centrality metrics
  • ClusteringHubStrength.csv Nodes sorted by hubs strength according to RQ1.3 for each used centrality
• Results Results gathered to answer RQ1-3
  • RQ1 Results for RQ1
    • ScaleFreeTest.csv Results of testing the degree distributions for Scale-free property (RQ1.1)
    • Kirkley.json Results of using four methods provided by Alec Kirkley [2] for hub detection in networks (RQ1.2)
    • ClusteringHubsAgreement.csv Nodes considered hubs by examining the scatter plots of centrality and clustering coefficient for each centrality (RQ1.3)
    • HubsAll.csv A dataframe of all microservice from all systems labelled as hub or non-hub by all approaches (including Arcan method, RQ1.4)
    • HubsTrue.csv A dataframe of all microservices from all systems labelled as hub by at least one approach
    • HubCounts.csv Count of microservice detect as hub for each approach
  • RQ2 Results for RQ2 (agreement between the methods)
    • HubJaccard_incoming.csv Jaccard coefficient of agreement between the methods analyzing the incoming connections
    • HubJaccard_outgoing.csv Jaccard coefficient of agreement between the methods analyzing the outgoing connections
    • HubJaccard_all.csv Jaccard coefficient of agreement between the methods analyzing both the incoming and outgoing connections
  • RQ3 Results for RQ3 (precision through manual validation)
    • ManualValidation.csv Nodes detected as hub by at least one approach labelled as True (Hub), Infra (Infrastructural hub), or False (non-Hub)
    • Precision.csv Precision of the hub detection approach calculated by considering IH as TP, FP, or ignoring them
• Scripts Scripts used to perform the study
  • centrality.py Compute all the centrality metrics and clustering coefficient as well as make the comparison figure of degree and degree centrality, and all SDG figures
  • scale-free-test.py Test the networks for the scale-free property by fitting a power law to the degree distribution
  • kirkley.py Perform hub detection with four approaches provided by Alec Kirkley [2]
  • clustering.py Compute the hub-like strength by subtracting the clustering coefficient from the centrality metric and make the scatter plots for RQ1.3
  • merge_results.py Combine the results of RQ1.1-RQ1.3, generate results for RQ1.4 and agreement and precision for RQ2 and RQ3
• Figures All the figures necessary for the work
  • SDGs SDGs of all studied systems
  • ClusteringScatter Scatter plots of centrality and clustering for RQ1.3
  • Comparison.pdf Comparison of CDFs of Degree and Degree centrality
  • ScaleFree Results of fitting a power law to the degree distributions
  • HubJaccard_incoming.pdf Jaccard coefficient of agreement between methods analyzing incoming connections
  • HubJaccard_outgoing.pdf Jaccard coefficient of agreement between methods analyzing outgoing connections
  • HubJaccard_all.pdf Jaccard coefficient of agreement between methods analyzing both the incoming and outgoing connections

References

[1] A. Bakhtin, M. Esposito, V. Lenarduzzi and D. Taibi, "Network Centrality as a New Perspective on Microservice Architecture," 2025 IEEE 22nd International Conference on Software Architecture (ICSA), Odense, Denmark, 2025, pp. 72-83, doi: 10.1109/ICSA65012.2025.00017. [2] Kirkley A. Identifying hubs in directed networks. Physical Review E. 2024 Mar;109(3):034310.