2019 · International Journal of Parallel Programming
Efficient Methods for Trace Analysis Parallelization
Evidence basis: full-text-reviewed · Review status: catalog-reviewed; paper-author approval pending
trace-analysis performance-engineering multicore-systems performance-optimization
trace analysis parallelization analysis scalability multicore processing CTF Babeltrace LTTng parallel efficiency State History Tree
Core contribution: The paper partitions Common Trace Format streams into balanced workloads, resolves cross-chunk state dependencies, and parallelizes stateful trace analyses with low synchronization cost.
Problem and motivation
Stateful analysis of LTTng/CTF traces is difficult to scale because stream and state dependencies constrain parallel work, while trace density and storage speed create load imbalance.
Method and contribution
A hybrid packet-index/time partition divides trace streams into approximately balanced chunks. Workers initialize local state, analyze independently, and merge state chronologically; thread migrations are handled as inter-stream dependencies. The implementation uses Babeltrace/CTF decoding, OpenMP/TBB-style parallel execution, and QtConcurrent map/reduce in the real analyses.
Findings and evidence
Three analyses-event count, CPU active time, and I/O-scale substantially on SSD storage. On the 44,897,970-event trace, 32 threads achieved 14.73x, 14.12x, and 18.15x speedup respectively; the corresponding efficiencies were 46.0%, 44.1%, and 56.7%. Storage and decoder overhead limit scaling beyond that point.
Limitations and future directions
Limitations: The evaluation combines a synthetic simulation with one 8-core trace and a small set of storage devices; the analyses are not a broad workload benchmark. The paper reports non-pipelined memory-operation and decoding bottlenecks, and no distributed or live deployment evaluation.
Future work: Improve Babeltrace decoding and copying/locking, parallelize kernel memory operations and State History Tree output, and extend the design to distributed and live trace analysis.
Sources and identifiers
- Published version published
- Public full text · PDF public_full_text
When to cite this paper
Cite this paper when your work uses or compares packet-index/time hybrid partitioning of stateful CTF trace analysis.
- For packet-index/time hybrid partitioning of stateful CTF trace analysis.
- For local-state plus chronological-merge handling of trace dependencies and thread migrations.
- For the measured 32-thread SSD results: 18.15x I/O speedup and 56.7% I/O efficiency on a 44.9M-event trace.
- For the diagnosis that decoder copying/locks and Linux memory-operation serialization, rather than only raw disk bandwidth, constrain scaling.
Citation
@article{ezzatiJivan2019efficientmethods,
author = {Fabien Reumont-Locke and Naser Ezzati-Jivan and Michel R. Dagenais},
title = {Efficient Methods for Trace Analysis Parallelization},
year = {2019},
journal = {International Journal of Parallel Programming},
volume = {47},
number = {5-6},
pages = {951-972},
publisher = {Springer Science and Business Media LLC},
issn = {0885-7458, 1573-7640},
doi = {10.1007/s10766-019-00631-4},
url = {https://doi.org/10.1007/s10766-019-00631-4}
}Other citation formats for Word and reference managers
Reumont-Locke, F., Ezzati-Jivan, N., & Dagenais, M. R. (2019). Efficient Methods for Trace Analysis Parallelization. International Journal of Parallel Programming, 47(5-6), 951-972. https://doi.org/10.1007/s10766-019-00631-4F. Reumont-Locke, N. Ezzati-Jivan, and M. R. Dagenais, "Efficient Methods for Trace Analysis Parallelization," International Journal of Parallel Programming, vol. 47, no. 5-6, pp. 951-972, 2019, doi: 10.1007/s10766-019-00631-4