This repository contains the source code for the paper "K2vTune: A Workload-aware Configuration Tuning for RocksDB" published on Information Processing & Management in January 2024. K2vTune is a workload-aware configuration tuning framework, which can recognize the configuration knobs of RocksDB according to the workload type and effectively consider multiple performance metrics using our knob2vec method.

• lifelines
• pytorch == 1.7.0
• python >= 3.8

• https://automl.github.io/SMAC3/main/index.html

conda install gxx_linux-64 gcc_linux-64 swig
pip install smac

target       : target workload number  
tf           : using teacher forcing, if not specify this, the model will be trained by non-teacher forcing  
train        : mode of train  
eval         : mode of train using pre-trained model(.pt)  
model_path   : if using eval mode, add pre-trained model path  
batch_size   : batch size for dataset
hidden_size  : hidden size of the model  
lr           : learning rate of the model
mode         : regression model type ['raw', 'dnn', 'gru', 'attngru']
attn_mode    : attention tyep ['dot', 'general', 'concat', 'bahdanau']
generation   : the counts of generation in Genetic Algorithm  
pool         : size of pool in genetic algorithm
optimization : choose optimization algorithm ['ga', 'smac']

• Training the model

Please modify arguments in the bash file. See train.sh

./main.sh train

or

python main.py --target ${target_idx} --tf --train --hidden_size ${hidden_size} --lr ${learning_rate} \
--generation ${generation_num} --pool ${pool_num}

• Training with pre-trained model path

Please modify arguments in the bash file. See eval.sh

./main.sh eval

or

python main.py --target ${target_idx} --tf --eval --model_path ${model_path} \
--generation ${generation_num} --pool ${pool_num}

We set the parameters as follows

• hidden_size = 128
• lr = 0.001
• generation = 100
• pool = 128
• mode = 'attngru'
• attn_mode = 'general'
• optimization = 'ga'