PRINS: Scalable Model Inference for Component-based System Logs

This repository contains the replication package for the paper titled "PRINS: Scalable Model Inference for Component-based System Logs", EMSE (2022). Please use the following citation when you want to cite our work:

Shin, D., Bianculli, D. & Briand, L. PRINS: scalable model inference for component-based system logs. Empir Software Eng 27, 87 (2022). https://doi.org/10.1007/s10664-021-10111-4

Authors

• Donghwan Shin ([email protected])
• Domenico Bianculli ([email protected])
• Lionel Briand ([email protected])

Prerequisites

• python 3.7+ (tested with 3.7 and 3.9)
• graphviz 10+ (tested with 10.0.1)
• JDK 8+ (for MINT, tested with openjdk 17.0.10)

Install

Firstly, to render generated models in PDF, you must install dot. Try the following command to check if dot is installed:

dot -V

If dot is not installed, you can install it by following this page. On Windows, you can follow the installation procedure.

Second, initialize python's virtual environment & install required packages:

python3 -m venv venv
source venv/bin/activate  # venv should be activated during the execution of PRINS
pip install -r requirements.txt

Finally, you must have JDK to run MINT, which is used as a backend for PRINS. Try the following command to check if JDK is installed:

java -version

If JDK is not installed, you can install it by following this page.

Dataset Check

You can simply run the following script:

(venv) python check_dataset_summary.py

It will save the summary data in dataset/dataset_summary.csv.

Unit Tests

You can run the unit tests:

cd PRINS
(venv) python -m unittest

All 50 tests should pass.

Execution Time Evaluation (RQ1, RQ2, RQ4)

(venv) PROMPT PRINS-expr % python run_model_inference.py -h
usage: run_model_inference.py [-h] [-s SYSTEM] [-n NUM_LOGS] [--prins_only]
                              [--mint_sys_only] [-d DUPLICATE_RANGE]
                              [-r REPETITIONS]

optional arguments:
  -h, --help            show this help message and exit
  -s SYSTEM, --system SYSTEM
                        System name (default=None)
  -n NUM_LOGS, --num_logs NUM_LOGS
                        Number of logs (default=all)
  --prins_only          Specify this to run PRINS only
  --mint_sys_only       Specify this to run MINT-SYS only
  -d DUPLICATE_RANGE, --duplicate_range DUPLICATE_RANGE
                        Input log duplication factor range 'from,to'
                        (default='1,1')
  -r REPETITIONS, --repetitions REPETITIONS
                        Number of repetitions (default=1)

Accuracy Evaluation (RQ3, RQ5)

(venv) PROMPT PRINS-expr % python run_k_folds_cv.py -h
usage: run_k_folds_cv.py [-h] [-k NUM_FOLDS] [-n NUM_LOGS] system technique

positional arguments:
  system                System name (e.g., Hadoop)

optional arguments:
  -h, --help            show this help message and exit
  -k NUM_FOLDS, --num_folds NUM_FOLDS
                        Number of folds
  -n NUM_LOGS, --num_logs NUM_LOGS
                        Number of logs (default: all)

Result Data Analysis

Run the script (note that the experimental results must be provided in /expr_output; by default, the results reported in the paper are given):

(venv) python run_analyze_results.py

The scripts will automatically generate the following files:

• rq1-boxplot.pdf
• rq2-boxplot.pdf
• rq3-boxplot.pdf
• rq4-line.csv
• rq5-table.csv
• rq5-table-size.csv
• message_distribution.csv

Licensing

PRINS is © 2021 University of Luxembourg and licensed under the GPLv3 license.

Please read PRINS licensing information.txt for details.