2021 · Wireless Communications and Mobile Computing
Debugging of Performance Degradation in Distributed Requests Handling Using Multilevel Trace Analysis
Evidence basis: full-text-reviewed · Review status: catalog-reviewed; paper-author approval pending
system-tracing microservices performance-analysis root-cause-analysis trace-analysis
distributed requests multilevel trace analysis LTTng Apache PHP MariaDB State History Tree cache hit ratio disk contention ApacheBench Trace Compass
Core contribution: The paper correlates LTTng traces from user space through kernel, storage, network, and multiple hosts in a disk-backed state model, enabling top-down diagnosis of distributed request latency.
Problem and motivation
Performance degradation in distributed requests can arise at user, kernel, storage, network, or remote-service layers, so single-layer tracing does not provide enough causal context.
Method and contribution
Instrument LAMP components and system layers with LTTng, synchronize hosts using causal network event pairs and convex-hull matching, maintain multilevel state trees, and perform top-down diagnosis from request latency to system call and kernel/storage/network states.
Findings and evidence
The case study localized application-level session/statistics work and disk queue contention from backup activity. The system distinguished cache-hit/miss latency classes and showed that line-level PHP tracing can approximately double response time, while selected UST tracing can approach the no-trace baseline.
Limitations and future directions
Limitations: The evaluation focuses on LAMP and selected instrumented cases; causal synchronization requires suitable event pairs and the available probes determine what can be explained. Network attacks, broader systems, and richer anomaly classes are not evaluated.
Future work: Add network-attack and host-anomaly analysis, broaden probes and workloads, and improve multilevel visualization and causal coverage.
Resources
Sources and identifiers
- Published version published
- Public Wiley PDF · PDF public_full_text
When to cite this paper
Cite this paper when your work uses or compares top-down multilevel root-cause analysis from distributed request latency to kernel/storage/network state.
- For top-down multilevel root-cause analysis from distributed request latency to kernel/storage/network state.
- For LTTng instrumentation spanning LAMP user space, system calls, kernel, filesystem, page cache, block layer, disk, and network.
- For cross-host causal synchronization using request/response event pairs and convex-hull matching.
- For the MediaWiki/WordPress session-check and `backup.sh` disk-queue case studies, including the tracing-cost boundary.
Citation
@article{ezzatiJivan2021debuggingof,
author = {Naser Ezzati-Jivan and Houssem Daoud and Michel R. Dagenais},
title = {Debugging of Performance Degradation in Distributed Requests Handling Using Multilevel Trace Analysis},
year = {2021},
journal = {Wireless Communications and Mobile Computing},
volume = {2021},
number = {1},
eid = {8478076},
publisher = {Wiley},
issn = {1530-8669, 1530-8677},
doi = {10.1155/2021/8478076},
url = {https://doi.org/10.1155/2021/8478076}
}Other citation formats for Word and reference managers
Ezzati-Jivan, N., Daoud, H., & Dagenais, M. R. (2021). Debugging of Performance Degradation in Distributed Requests Handling Using Multilevel Trace Analysis. Wireless Communications and Mobile Computing, 2021(1), 8478076. https://doi.org/10.1155/2021/8478076N. Ezzati-Jivan, H. Daoud, and M. R. Dagenais, "Debugging of Performance Degradation in Distributed Requests Handling Using Multilevel Trace Analysis," Wireless Communications and Mobile Computing, vol. 2021, no. 1, 2021, doi: 10.1155/2021/8478076