Enhanced Execution Trace Abstraction Approach Using Social Network Analysis Methods

Catalog abstract summary

This paper applies social-network-analysis techniques to system execution traces. The workflow collects Linux kernel and user-space traces with LTTng, constructs a directed weighted graph of thread interactions, detects communities with the Louvain method, and ranks important threads with PageRank. Two use cases address trace filtering and virtual-machine clustering. In the reported evaluation, the analysis adds a 5.3% slowdown to the traced program, extracts 509 threads and 6,015 interactions from a 316 MB trace, and computes graph metrics in 1,599 ms.

Source: public PDF at fb-swt.gi.de, downloaded and PDF-signature verified 2026-08-03

Problem and motivation

Large execution traces expose many thread interactions, making it difficult to retain a concise, useful view while filtering irrelevant activity.

Method and contribution

Collect LTTng kernel events, construct a weighted directed thread-interaction graph, detect communities with Louvain, rank important nodes with within-community PageRank, and apply the resulting abstraction/filter through Trace Compass EASE.

Findings and evidence

On the reported Ubuntu/LTTng trace, the method processed 509 threads and 6,015 distinct interactions from a 316 MB trace, with 1,599 ms extraction time and 5.3% overall slowdown. The paper demonstrates trace filtering; VM clustering is only an abstract-level claim in this evidence boundary.

Limitations and future directions

Limitations: This is a three-page short paper with one small evaluation, no systematic abstraction baseline, and no reported filtering-quality metric. Workload details, parameter sensitivity, and a VM-clustering evaluation are unknown.

Future work: Validate the abstraction across workloads, compare filtering baselines, quantify information loss/usefulness, and provide the missing VM-clustering method and evaluation if that use case is retained.

Sources and identifiers

• Public PDF public_full_text
• DBLP record metadata_record

When to cite this paper

Cite this paper when your work uses or compares applying Louvain community detection to thread-interaction graphs for trace abstraction.

• For applying Louvain community detection to thread-interaction graphs for trace abstraction.
• For PageRank-based prioritization of important threads inside interaction communities.
• For a Trace Compass EASE script that turns the graph abstraction into a global trace filter.
• For the concrete 5.3% slowdown, 316 MB trace, 509-thread, and 1,599 ms extraction measurements; not for a validated VM-clustering result.

Citation

@article{ezzatiJivan2020enhancedexecution,
  author = {Ji Wang and Naser Ezzati-Jivan},
  title = {Enhanced Execution Trace Abstraction Approach Using Social Network Analysis Methods},
  year = {2020},
  journal = {Softwaretechnik-Trends},
  volume = {40},
  pages = {58-60},
  url = {https://fb-swt.gi.de/fileadmin/FB/SWT/Softwaretechnik-Trends/Verzeichnis/Band_40_Heft_3/SSP2020_Wang.pdf}
}

Wang, J., & Ezzati-Jivan, N. (2020). Enhanced Execution Trace Abstraction Approach Using Social Network Analysis Methods. Softwaretechnik-Trends, 40, 58-60. https://fb-swt.gi.de/fileadmin/FB/SWT/Softwaretechnik-Trends/Verzeichnis/Band_40_Heft_3/SSP2020_Wang.pdf

J. Wang and N. Ezzati-Jivan, "Enhanced Execution Trace Abstraction Approach Using Social Network Analysis Methods," Softwaretechnik-Trends, vol. 40, pp. 58-60, 2020, [Online]. Available: https://fb-swt.gi.de/fileadmin/FB/SWT/Softwaretechnik-Trends/Verzeichnis/Band_40_Heft_3/SSP2020_Wang.pdf