Resource Life-Cycle Aware Noise Detection via Kernel Event Monitoring

Catalog abstract summary

The available synopsis describes passive kernel-event monitoring and alignment of CPU, disk, and network request life cycles for noise detection; detailed evaluation was not captured.

Source: OpenAlex abstract metadata and public author synopsis reviewed; publisher abstract not reproduced because reuse permission was not established.

Problem and motivation

Noise can originate in different phases of a resource request, and intrusive instrumentation can perturb the execution being measured. Aggregate CPU, disk, or network values often show that a disturbance exists but not which request phase or event caused it.

Method and contribution

The study uses LTTng kernel events and defines request life-cycle phases with request/response relationships. Disk phases include block wait from issue to insert and disk wait from completion to issue; network phases include transmit and receive waits. Trace Compass 8.3 visualizes the aligned states. The Ubuntu 22.04/Intel i7 3.6 GHz/32 GB environment runs in a VirtualBox 6.1 guest and uses wget/curl, Python disk activity, grep, sysbench prime workloads, and OS noise scenarios to exercise CPU, disk, and network paths.

Findings and evidence

The case studies show that life-cycle alignment distinguishes transient network fluctuation, disk-wait behavior, CPU scheduling, and interrupt-related disturbances. The paper reports disk-event counts below 8.950 in one condition versus approximately 41.935–41.940 and 47 in others, and identifies an interrupt episode of about 0.6 seconds for TID 8615 on CPU 3. Turning CPU noise on/off changes the visible request-phase pattern, illustrating why phase-aware evidence is useful.

Limitations and future directions

Limitations: The validation uses simulated workloads and selected thresholds/event definitions; it does not establish detector accuracy against a broad ground-truth corpus or production deployments. Event collection overhead, portability to other kernels, and sensitivity to incomplete request/response matching require further study.

Future work: Formalize phase-aware labels and thresholds, test production and distributed traces, quantify precision/recall and detection delay, evaluate tracing overhead, and support automatic adaptation to new resource-event schemas.

Sources and identifiers

• Published version published
• CASCON program public_source_record

When to cite this paper

Cite this paper when localizing performance noise to CPU, disk, or network request-life-cycle phases.

• Passive LTTng event collection aligned to request/response phases.
• Disk and network wait-state definitions.
• Trace Compass visual analysis of injected resource noise.
• The simulated-workload and threshold/ground-truth limitations.

Citation

@inproceedings{ezzatiJivan2024resourcelife,
  author = {Morteza Noferesti and Ben Grandy and Naser Ezzati-Jivan},
  title = {Resource Life-Cycle Aware Noise Detection via Kernel Event Monitoring},
  year = {2024},
  booktitle = {2024 IEEE International Conference on Collaborative Advances in Software and COmputiNg (CASCON)},
  pages = {1-10},
  publisher = {IEEE},
  doi = {10.1109/CASCON62161.2024.10838205},
  url = {https://doi.org/10.1109/CASCON62161.2024.10838205}
}

Noferesti, M., Grandy, B., & Ezzati-Jivan, N. (2024). Resource Life-Cycle Aware Noise Detection via Kernel Event Monitoring. In 2024 IEEE International Conference on Collaborative Advances in Software and COmputiNg (CASCON) (pp. 1-10). https://doi.org/10.1109/CASCON62161.2024.10838205

M. Noferesti, B. Grandy, and N. Ezzati-Jivan, "Resource Life-Cycle Aware Noise Detection via Kernel Event Monitoring," in 2024 IEEE International Conference on Collaborative Advances in Software and COmputiNg (CASCON), pp. 1-10, 2024, doi: 10.1109/CASCON62161.2024.10838205