Space-Time Analysis of Network Alarms
Detecting causal relationships in a system log file is often complex, as such files are huge and noisy. In this project, we designed a new data structure, called DIG-DAG, that stores chains of alarms observed in a log file. Once fed, it can be queried to isolate and display faulty alarm patterns.
Automatic Log Parsing
Automatic log parsing transforms unstructured data (logs & command outputs) into a structured format. This is the first step towards automatic log analysis tasks, enabling efficient usage analysis, anomaly detection, failure diagnosis, and performance monitoring.
Making Algorithms Vote
Many problems, for example in decision making, cannot be solved exactly and use heuristic algorithms that assign scores to the various choices available. We propose to aggregate the scores in a way that uses correlations between algorithms rather than suffering from them. It relies on a Singular-Value Decomposition (SVD) and the Nash Product of utilities. It is more efficient than trivial aggregation in terms of finding the ground truth, and it is more robust than an approach based on maximum likelihood.
Our aggregation method (SVD-Nash) does not need any training, but it performs as well as trained Maximum Likelihood (ML-1000)
APA style publications
-
Maxime Raynal, Marc-Olivier Buob, Georges Quénot, "A novel pattern-based edit distance for automatic log parsing", 2022 26th International Conference on Pattern Recognition (ICPR), 2022
-
Maxime Raynal, Marc-Olivier Buob, Georges Quénot, " A novel pattern-based edit distance for automatic log parsing: Implementation and reproducibility notes", RRPR 2022 : Fourth Workshop on Reproducible Research in Pattern Recognition (satellite event of ICPR 2022) [pdf]
-
Théo Delemazure, François Durand, Fabien Mathieu. Poster “Election with dependent voters”. Workshop on Computational Social Choice (COMSOC-2021), 2021. [pdf] Best Student Poster Award
-
Théo Delemazure, François Durand, Fabien Mathieu. Démocratie à géométrie variable (à l'usage des algorithmes). ALGOTEL 2021 - 23èmes Rencontres Francophones sur les Aspects Algorithmiques des Télécommunications, Jun 2021, La Rochelle, France. [pdf]
-
Achille Salaün, Anne Bouillard, Marc-Olivier Buob, "Demo Abstract: End-to-end Root Cause Analysis of a Mobile Network," IEEE INFOCOM 2020 - IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), 2020, pp. 1324-1325, doi: 10.1109/INFOCOMWKSHPS50562.2020.9162999. [pdf]
-
Anne Bouillard, Achille Salaün, Marc-Olivier Buob, " Space-time pattern extraction in alarm logs for network diagnosis", 2019 Machine Learning for Networking (MLN), 2019, pp. 134-153. [pdf]
-
Anne Bouillard, Marc-Olivier Buob, Achille Salaün, Maxime Raynal, "Log Analysis via Space-time Pattern Matching," 2018 14th International Conference on Network and Service Management (CNSM), 2018, pp. 303-307. [pdf]
-
Anne Bouillard, Marc-Olivier Buob, Maxime Raynal, "DIG-DAG construction for root-cause analysis" [patent]
Discover more
Biographies
Location
Service
Awards