DBTune is a customized and efficient database tuning system that can automatically find good configuration knobs for a database system. It supports multiple tuning scenarios, including performance tuning, resource-oriented tuning or multiple-objective tuning defined by the users. DBTune is equipped with state-of-the-art techniques for tuning a database. DBTune is designed and developed by the database team from the DAIR Lab at Peking University.
-
Optimized for database tuning: DBTune has customized functions and user-friendly interfaces for tuning the DBMSs. Users can conduct performance tuning, resource tuning or multiple-objective tuning by simply setting their tuning objectives.
-
Comprehensive technique supports and automatic selection: DBTune is equipped with state-of-the-art tuning techniques. It automatically chooses the proper algorithms for a specific tuning task to separate algorithm selection complexity away from the user.
-
Performance boosted by transfer learning: DBTune extracts knowledge from historical tuning tasks to speed up the current tuning. The more users utilize DBTune, the faster the tuning will be.
DBTune supports the whole pipeline of configuration tuning, including knob selection, configuration tuning and knowledge transfer. Each module is equipped with multiple algorithm choices. For a given tuning task, DBTune automatically selects a proper solution path among the choices.
Installation Requirements:
- Python >= 3.6
To install the newest DBTune package, type the following scripts on the command line:
git clone git@github.com:Blairruc-pku/DBTuner.git && cd DBTune
pip install -r requirements.txt
pip install .DBTune currently supports three database benchmarks: OLTP-Bench, SYSBENCH and JOB. Please reffer to the details instuction for preparing the workloads.
To provide the database connection information, the users need to edit the config_auto.ini.
db = mysql
host = 127.0.0.1
port = 3306
user = root
passwd =DBTune currently supports to be deployed on MySQL and PostgreSQL using an integrated framework. It provides several settings for database connections, including remote/local database connection, tuning non-dynamic knobs with restarts, and tuning with resource isolation. Please reffer to the details configurations for more information.
-
Specify the tuning objective in
config_auto.ini. Here are some examples.Performance tuning, e.g., maximizing throughputs.
task_id = performance1 performance_metric = ['tps']
Resource-oriented tuning, e.g., minimizing cpu resource while throughputs > 200 txn/s and 95th percentile latency < 60 sec.
task_id = resource1 performance_metric = ['-cpu'] #constraints: Non-positive constraint values (”<=0”) imply feasibility. constraints = ["200 - tps", "latency - 60"]
Multiple objective tuning, e.g., maximizing throughput and minimizing I/O.
task_id = mutiple1 performance_metric = ['tps', '-cpu] reference_point = [0, 100]
-
Conduct Tuning.
cd scripts python optimize.py --config=config_performance.iniFor more information, please refer to the specific tuning settings .
Facilitating Database Tuning with Hyper-Parameter Optimization: A Comprehensive Experimental Evaluation[PDF]
Xinyi Zhang, Zhuo Chang, Yang Li, Hong Wu, Jian Tan, Feifei Li, Bin Cui.
The 48th International Conference on Very Large Data Bases .
VLDB 2022, CCF-A
Towards Dynamic and Safe Configuration Tuning for Cloud Databases[PDF]
Xinyi Zhang, Hong Wu, Yang Li, Jian Tan, Feifei Li, Bin Cui.
ACM Conference on Management of Data .
SIGMOD 2022, CCF-A
ResTune: Resource Oriented Tuning Boosted by Meta-Learning
for Cloud Databases[PDF]
Xinyi Zhang, Hong Wu, Zhuo Chang, Shuowei Jin, Jian Tan, Feifei Li,
Tieying Zhang, and Bin Cui.
ACM Conference on Management of Data .
SIGMOD 2021, CCF-A
If you have any technical questions, please submit new issues.
If you have any other questions, please contact Xinyi Zhang[zhang_xinyi@pku.edu.cn] and Zhuo Chang[z.chang@pku.edu.cn].
The entire codebase is under MIT license.