Efficient Heap Monitoring Tool for Memory Leak Detection and Root-cause Analysis

Problem and motivation

Long-lived C/C++ services can accumulate leaked allocations gradually, while repeated heap snapshots and tools such as Valgrind, ASAN, or LSAN can impose high overhead or require a special build and end-of-run analysis. Developers also need a causal view from leaked bytes to allocation sites.

Method and contribution

An LD_PRELOAD library intercepts malloc, calloc, realloc, and free. A single heap-walker thread periodically gathers lightweight statistics; allocation headers retain call-stack information, and a shared-memory command-line interface controls the monitor. The Minimum Envelope filter removes transient jumps and dips to isolate persistent upward active-heap trends in bytes or block counts. JSON Google Trace Event output is imported into Trace Compass and correlated with LTTng kernel events; Sankey and call-graph views aggregate allocation stacks, while age-based and minimum-envelope views expose persistent allocations.

Findings and evidence

The paper reports 5–12% CPU overhead and less than 10% heap-memory impact in its stated implementation. Synthetic leak probabilities of 0.1, 0.01, 0.001, and 0.0001 over 60-minute runs converge within 10% of the nominal rate, with approximately 91% precision in the reported evaluation. Industrial nested-call, binary-search-tree, and linked-list cases illustrate root-cause localization; at a 0.001 leak probability, Valgrind leak-check=full is reported at roughly six times native CPU cost.

Limitations and future directions

Limitations: The quantitative evidence combines synthetic leak injection with a limited industrial example, and threshold selection affects persistence detection and false positives. The paper calls for 90–100 additional runs to strengthen precision estimates and notes that source-file and line rendering in the call graph needs improvement.

Future work: Run larger production-like studies, quantify recall and time-to-detection under allocator and concurrency variation, improve source mapping and threshold calibration, and evaluate false-positive control when legitimate caches or workload phases grow the heap persistently.

Sources and identifiers

• Published version published
• DBLP record public_source_record

When to cite this paper

Cite this paper when your work monitors heap growth with low-intrusion allocation interposition and call-stack root-cause views.

• LD_PRELOAD interception of allocation/free operations and a single heap walker.
• Minimum Envelope filtering for persistent heap-growth detection.
• Trace Compass/LTTng correlation and Sankey/call-graph leak localization.
• The reported overhead and synthetic/industrial validation boundary.

Citation

@inproceedings{ezzatiJivan2021efficientheap,
  author = {Vahid Azhari and Simar Bhamra and Naser Ezzati-Jivan and Francois Tetreault},
  title = {Efficient Heap Monitoring Tool for Memory Leak Detection and Root-cause Analysis},
  year = {2021},
  booktitle = {2021 IEEE International Conference on Big Data (BigData)},
  pages = {3020-3030},
  publisher = {IEEE},
  doi = {10.1109/BigData52589.2021.9671473},
  url = {https://doi.org/10.1109/BigData52589.2021.9671473}
}

Azhari, V., Bhamra, S., Ezzati-Jivan, N., & Tetreault, F. (2021). Efficient Heap Monitoring Tool for Memory Leak Detection and Root-cause Analysis. In 2021 IEEE International Conference on Big Data (BigData) (pp. 3020-3030). https://doi.org/10.1109/BigData52589.2021.9671473

V. Azhari, S. Bhamra, N. Ezzati-Jivan, and F. Tetreault, "Efficient Heap Monitoring Tool for Memory Leak Detection and Root-cause Analysis," in 2021 IEEE International Conference on Big Data (BigData), pp. 3020-3030, 2021, doi: 10.1109/BigData52589.2021.9671473