BQSched & BQSched⁺

Here are the codes for seven different experiments as follows on TPC-DS benchmark with PostgreSQL:

• train_from_scratch (BQSched)
• train_on_clusters (BQSched)
• pre_train (BQSched)
• fine_tune (BQSched)
• train_from_scratch_on_varying (BQSched⁺)
• pre_train_on_varying (BQSched⁺)
• fine_tune_on_varying (BQSched⁺)

Quick Start

Please follow the steps below to run the script for each mode:

1. Set up the Python 3.8.16 environment with pip install -r requirements.txt.
2. Replace the files in the installed packages with the corresponding files in the replace/ folder, where replacing - with / in the filenames gives you the paths to the target files to replace, starting from the site-packages/ folder in your own environment.
3. Run the script python -u train.py --mode {mode_name}, where you can run different modes by specifying different mode_name values.
   • The choices for mode_name are ['train_from_scratch', 'train_on_clusters', 'pre_train', 'fine_tune', 'train_from_scratch_on_varying', 'pre_train_on_varying', 'fine_tune_on_varying'].
   • The corresponding configurations for each mode are in files modes/args/{mode_name}.py and modes/config/{mode_name}.ini.
   • Note that you need to replace the configurations for the underlying DBMS connection with information about your target DBMS (including host, port, user, password, max_worker, etc.), and collect workload information in advance with reference to the contents in the folder envs/scheduler/cache/your_host/.
4. Check the runtime logs in the file outs/{host_postfix}.runtime.out and the results in the logs/ folder.

Reproducing Key Results

Coming soon.

Citing

Coming soon.