2023 · Companion of the 2023 ACM/SPEC International Conference on Performance Engineering (ICPE '23 Companion)

Software Mining - Investigating Correlation between Source Code Features and Michrobenchmark's Steady State

Amirmahdi Khosravi Tabrizi | Naser Ezzati-Jivan

Evidence basis: full-text-reviewed · Review status: catalog-reviewed; paper-author approval pending

performance-analysis performance-engineering machine-learning

Java Microbenchmark Harness JMH srcML Lizard Apriori Java microbenchmarks JVM warmup steady state source-code features BMNFC BMLOC DNNSF

Core contribution: The study examines whether static source-code features are associated with the steady-state behavior of Java microbenchmarks during JVM warmup.

Problem and motivation

JVM JIT compilation makes Java microbenchmark results unstable during warmup; the study asks whether static source features are associated with reaching steady state (p. 1).

Method and contribution

The study uses 586 JMH results from 30 open-source Java projects and 10 forks per benchmark method. srcML and Lizard extract 11 static features; dynamic features include DNNSF, DANCP, and DANITS. Twenty-seven records were excluded after feature-extraction errors. Manual comparisons at benchmark-method/file level are followed by Apriori descriptive association mining (pp. 2-4).

Findings and evidence

At benchmark-method level, 321 cases are steady, 236 inconsistent, and 2 non-steady. Reported Apriori associations are BMNFC_HIGH -> DNNSF_ZERO with confidence 0.63 and BMLOC_HIGH -> DNNSF_ZERO with confidence 0.61; the paper interprets higher function count/LOC as associated with zero non-steady forks (p. 4).

Limitations and future directions

Limitations: This is exploratory association analysis, not causal evidence. Only two benchmark methods are in the non-steady class, 27 records were removed, and independent replication/generalization across JVMs, hardware, languages, and workloads are not established. The paper does not specify a complete OS/hardware/JVM configuration.

Future work: Add source features for system calls, threading, locks, and other function-call types (p. 4).

Resources

Sources and identifiers

When to cite this paper

Cite this paper when your work uses or compares exploratory mining of source-code LOC/function-count features against JMH steady-state indicators.

Citation

BibTeX
@inproceedings{ezzatiJivan2023softwaremining,
  author = {Amirmahdi Khosravi Tabrizi and Naser Ezzati-Jivan},
  title = {Software Mining - Investigating Correlation between Source Code Features and Michrobenchmark's Steady State},
  year = {2023},
  booktitle = {Companion of the 2023 ACM/SPEC International Conference on Performance Engineering (ICPE '23 Companion)},
  pages = {107-111},
  publisher = {ACM},
  doi = {10.1145/3578245.3584695},
  url = {https://doi.org/10.1145/3578245.3584695}
}
Other citation formats for Word and reference managers
APA 7
Tabrizi, A. K., & Ezzati-Jivan, N. (2023). Software Mining - Investigating Correlation between Source Code Features and Michrobenchmark's Steady State. In Companion of the 2023 ACM/SPEC International Conference on Performance Engineering (ICPE '23 Companion) (pp. 107-111). https://doi.org/10.1145/3578245.3584695
IEEE
A. K. Tabrizi and N. Ezzati-Jivan, "Software Mining - Investigating Correlation between Source Code Features and Michrobenchmark's Steady State," in Companion of the 2023 ACM/SPEC International Conference on Performance Engineering (ICPE '23 Companion), pp. 107-111, 2023, doi: 10.1145/3578245.3584695

Readable Markdown record · JSON record · Download RIS