Before, running any experimental code, we must first create a supplemental script for dropping Linux's file system cache.
Go into the drop_caches directory, run make, update owner and set the sticky bit:
cd drop_caches
make
sudo chown root drop_caches
sudo chmod u+s drop_caches
Then, install the binary somewhere and add to PATH to ensure the experiment scripts can use it.
We use a set of python scripts to generate and import data into a compatible format. These scripts require the following python packages:
- numpy
- pandas
- scikit-learn
- click
First, download the original csv from here.
Then, use scripts/gen_forest_data.py to increase the dataset size, import the data into a compatible format, and generate the set of all possible forest predicates.
python scripts/gen_forest_data.py <input-csv> <output-db-path> <output-preds-path>
Use the scripts scripts/tbl2db.py and scripts/tpc-h-tbl2db.py to import .tbl data from the CH-benchmark and TCP-H respectively:
python scripts/tbl2db.py <input-ch-data-dir> <output-db-path>
python scripts/tpc-h-tbl2db.py <input-tpc-h-data-dir> <output-db-path>
Both scripts expect a single .tbl file containing all the data.
Run the forest experiments with:
cargo build --release
./target/release/run-forest-exp --preds <forest-preds-path> <db-path>
The number of randomly generated queries and the number of trials per query can be controlled with -n and -t respectively.
Run these experiments:
cargo build --release
./target/release/tpc-h-exp --db-path <db-path>
./target/release/ch-exp --db-path <db-path>
The number of trials can be controlled with -t.
To run tests, simply run:
cargo test