This repository includes the artifacts (including data and source code) of the our paper: "Multi-Intention Aware Configuration Selection for Performance Tuning"
The repository includes the following artifacts:
dataset: Section 2. Understanding Side-effects- labeled dataset including 7,325 parameters from 13 software covering four software domains
- labeled dataset including 735 parameters from PostgreSQL, Squid, Spark (in RQ1)
expansion: Section 3. Semi-supervised Data Expansion- source code
- small-scaled labeled data (1,292 parameters) to be expanded
- rules mined and new data expanded at each iteration
- domain-specific synonym list retrived from the study in Section 2 and supplemented from wiki.
- results including how many & how accurate are the new data
model: Section 4. Learning Based Model to Predict Tuning Guidance- source code
- training data (24,528 pieces, obtained after expansion)
- testing data (735 parameters from PostgreSQL, Squid, Spark)
comparing_existing: RQ2. Comparing SafeTune with State-of-art-tool- scripts and commands to validate the parameters missed by the existing work do have performance impacts
- results including
- performance impact of each parameter
- FULL comparision between SafeTune and the existing work
- all the inital testing results, by which we get the evaluation result in RQ2
case_study: RQ3. Effectiveness of SafeTune in Helping OtterTune- the other four cases that are not present in the paper due to the limited space