This project includes the different modules to demonstrate our work on auto-scaling Redis nodes based on its workload.
To setup the infrastructure follow the instructions inside the vcl-setup, redis-docker, scale-node-docker, workload-docker folders.
To install the prediction engine follow the instructions inside the prediction folder.
The current system only supports Azure, so you need azure credentials for things to work properly.
The prediction module consists of two models:
- ARIMA (Click here to see the code)
- RNN (Click here to see the code)
Folder structure (Details only about important files):
Inside prediction folder
.
├── rnn # Python code files of RNN
├── arima # Python code files of ARIMA
├── rnn_model.sav # Pre-trained model used in online-phase
├── predictionEngine.py # Main python file which calls consumer for streaming data and calls RNN and ARIMA to predict
├── consumer.py # Fetches data from Kafka and gives the response to predictionEngine
├── forecast.pickle # A trained ARIMA model saved using pickle
└── README.md
Inside RNN folder
.
├── multivariate_rnn-offline.py # Offline phase of RNN which trains the model
├── multivariate_rnn.py # Considers latency,memory and CPU for prediction
├── online_rnn.py # Python file that runs during the online-phase(integrated with system)
├── with_window.py # RNN with one node varying window size(Failed method 01)
├── multivariate_single_node.py # RNN with one node(Failed method 01)
└── README.md # Code described in detail for RNN
Inside ARIMA folder
.
├── auto_arima.py # AutoARIMA which sets value of p,q,d by choosing lowest AIC values
├── arima.py # ARIMA Function which chooses values of p,q from ACF & PACF,d from ch-test
└── README.md # Code described in detail for ARIMA