Hugging Face's logo

Datasets:
logfit-project
/
BGL

Tasks:
Text Classification
Modalities:
Tabular
Text
Formats:
parquet
Languages:
English
Size:
1M - 10M
Libraries:
Datasets
Dask
License:
Dataset card Data Studio Files
xet
Community
BGL / README.md
calmodovar's picture
calmodovar
Update dataset card
09a2641 verified
preview code
|
raw
history blame contribute delete
2.51 kB
metadata
pretty_name: BGL
dataset_name: logfit-project/BGL
task_categories:
  - text-classification
language:
  - en
size_categories:
  - 1M<n<10M
annotations_creators:
  - logfit-project
license: other

Dataset Card for logfit-project/BGL

Dataset Summary

The BlueGene/L (BGL) dataset contains console logs emitted by a 131,072-processor BlueGene/L supercomputer operated at Lawrence Livermore National Laboratory. Each line records hardware or software events tagged with an alert category, enabling downstream research on alert detection, prediction, and log analytics.

Supported Tasks and Leaderboards

  • anomaly-detection: binary or multi-class classification of alert categories.
  • log-parsing: template discovery and sequence modeling for high-performance computing (HPC) systems.

Dataset Structure

  • label: Alert category tag (- indicates a non-alert informational message).
  • timestamp: Unix epoch timestamp associated with the log event.
  • date: Calendar date formatted as YYYY.MM.DD.
  • node: Hardware node identifier that emitted the log line.
  • time: Precise timestamp including microseconds (YYYY-MM-DD-HH.MM.SS.xxxxxx).
  • node_repeat: Repeated node identifier found in the structured dataset.
  • type: High-level event type (e.g., RAS).
  • component: Subsystem reporting the log (e.g., KERNEL).
  • level: Severity level accompanying the event.
  • content: Verbose description of the underlying event.
  • anomaly: Binary indicator (1 for alert labels, 0 for non-alert -).

Source Data

Dataset Creation

The raw BlueGene/L logs are parsed with a deterministic regular expression to reproduce the schema published in BGL_2k.log_structured.csv. The transformation streams lines sequentially so the full corpus can be processed without loading the entire file into memory.

Uses

Suitable for supervised alert detection, failure prediction, anomaly detection, sequence modeling of HPC logs, and benchmarking log parsing techniques.

Citation

  • Adam J. Oliner, Jon Stearley. "What Supercomputers Say: A Study of Five System Logs", DSN 2007.
  • Jieming Zhu, Shilin He, Pinjia He, Jinyang Liu, Michael R. Lyu. "Loghub: A Large Collection of System Log Datasets for AI-driven Log Analytics", ISSRE 2023.

Dataset Statistics

  • Number of log lines: 4747963