2023 · IEEE International Conference on Big Data
AltOOM: A Data-driven Out of Memory Root Cause Identification Strategy
Evidence basis: metadata-only · Review status: catalog-reviewed; paper-author approval pending
resource-analysis root-cause-analysis system-tracing predictive-monitoring
out-of-memory OOM diagnosis data-driven RCA resource analysis memory pressure forecasting process-level profiling Perf sar
Core contribution: AltOOM combines early memory-pressure forecasting with selective process-level profiling to identify the process most responsible for an impending out-of-memory event.
Problem and motivation
Linux's reactive OOM killer can terminate a high-memory process even when a lower-memory process is the one whose memory usage is growing; resource-constrained systems need earlier warning and lower-overhead process attribution.
Method and contribution
AltOOM samples 34 system-level signals at 0.5-second intervals - 28 virtual-memory statistics, three memory-related system calls (brk, sbrk, and mmap), and three kernel events (kmalloc, mm_page_alloc, and vmscan) - using perf and sar. It labels pressure at %memused >= 85%, compares SVM, vanilla DNN, and bidirectional-LSTM predictors, filters 34 features to 15, then uses burst-collected process-level allocation signals and moving-average growth ranking after an alert.
Findings and evidence
In the reported evaluation, the feature-filtered DNN reaches 0.82 accuracy for (n,k)=(4,3) and 0.81 for (3,3); the abstract reports 85% memory-pressure forecasting accuracy. AltOOM process identification reaches 0.56-0.83 as the burst count increases from 3 to 7, versus 0.42 for the Linux OOM killer. A Firefox PDF-preview case reports 0.83 and 0.79 forecasting accuracy for (4,3) and (3,3), respectively.
Limitations and future directions
Limitations: The method can miss gradual memory buildup when only three timestamps (1.5 seconds) are observed. Fixed-rate monitoring creates overhead, and the evaluation centers on generated pressure scenarios plus a Firefox case rather than broad production-device coverage.
Future work: Evaluate adaptive sampling of rates, metrics, and metric groups, and implement actions such as controlling or adjusting the responsible process, terminating it, or restarting the system.
Sources and identifiers
- Published version published
When to cite this paper
Cite this paper when studying proactive memory-pressure forecasting or process-level out-of-memory root-cause identification.
- Perf/sar-based collection of VM statistics, memory-related system calls, and kernel events for memory-pressure prediction.
- Burst-based process profiling and moving-average ranking for identifying the responsible process.
- Comparison with the Linux OOM killer and reported forecasting and process-identification accuracy in simulated and Firefox cases.
Citation
@inproceedings{ezzatiJivan2023altooma,
author = {Pranjal Chakraborty and Naser Ezzati-Jivan and Vahid Azhari and François Tetreault},
title = {AltOOM: A Data-driven Out of Memory Root Cause Identification Strategy},
year = {2023},
booktitle = {IEEE International Conference on Big Data},
pages = {1637-1646},
publisher = {IEEE},
doi = {10.1109/bigdata59044.2023.10386937},
url = {https://doi.org/10.1109/bigdata59044.2023.10386937}
}Other citation formats for Word and reference managers
Chakraborty, P., Ezzati-Jivan, N., Azhari, V., & Tetreault, F. (2023). AltOOM: A Data-driven Out of Memory Root Cause Identification Strategy. In IEEE International Conference on Big Data (pp. 1637-1646). https://doi.org/10.1109/bigdata59044.2023.10386937P. Chakraborty, N. Ezzati-Jivan, V. Azhari, and F. Tetreault, "AltOOM: A Data-driven Out of Memory Root Cause Identification Strategy," in IEEE International Conference on Big Data, pp. 1637-1646, 2023, doi: 10.1109/bigdata59044.2023.10386937