Failure-Dataset-OpenStack

This failure dataset contains information on the events collected in the OpenStack cloud computing platform during three different campaigns of fault-injection experiments performed with three different workloads (DEPL, NET, and STO).

The dataset contains three folders referring to the three different fault-injection campaigns (one per each workload). In every folder, you can find:

• SEQ.tsv: a matrix with N rows and d columns containing the events collected during the fault-injection experiments. N is the number of fault-injection experiments, d represents the number of different event types (unique events).
• LCS_with_VMM.tsv: a matrix with N rows and 2d columns containing the anomalies identified by the LCS with VMM approach during the fault-injection experiments. N is the number of fault-injection experiments, d represents the number of different event types (unique events). The first d columns (1 to d) represent the number of spurious anomalies, the last d columns (d+1 to 2d) represents the number of missing anomalies.
• Failure_Labels.txt: a file containing the label class associated with every fault-injection experiment. The class label is expressed by a number.

The number N of experiments and the number d of events depend on the workload used during the fault-injection experiments. Therefore, the matrix dimension in the tsv files change in every folder. We invite the reader to find further information on the data representation, the workloads, the anomaly types, and the failure classes by reading the paper Enhancing the Analysis of Software Failuresin Cloud Computing Systems with Deep Learning accepted at the Journal of Systems and Software:

@article{cotroneo2021enhancing,
  title={Enhancing the analysis of software failures in cloud computing systems with deep learning},
  author={Cotroneo, Domenico and De Simone, Luigi and Liguori, Pietro and Natella, Roberto},
  journal={Journal of Systems and Software},
  volume={181},
  pages={111043},
  year={2021},
  publisher={Elsevier}
}

The slide presentation of the paper is available on slideshare, while you can find the video presentation of the paper on Youtube.

Failure Classes

In our fault-injection experiments, we found different failure modes. We associated to each failure-mode a numeric value (Failure Label).

DEPL

In the fault-injection campaing performed with the DEPL workload, the experiments experienced 6 different types of failures:

• Failure Label 1: Instance Failure
• Failure Label 2: Volume Failure
• Failure Label 3: SSH Failure
• Failure Label 4: Cleanup Failure
• Failure Label 5: Network Failure
• Failure Label 6: No Failure

NET

In the fault-injection campaign performed with the NET workload, the experiments experienced 4 different types of failures:

• Failure Label 1: Instance Failure
• Failure Label 2: Network Failure
• Failure Label 3: SSH Failure
• Failure Label 4: No Failure

STO

In the fault-injection campaign performed with the DEPL workload, the experiments experienced 4 different types of failures:

• Failure Label 1: Volume Failure
• Failure Label 2: Instance Failure
• Failure Label 3: Cleanup Failure
• Failure Label 4: No Failure