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 e 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.