README.md

AI_Component

The components of the AI part (as shown in the diagram below) are containerized and available at the following repositories:

• https://quay.io/repository/mdeloche/power10-mma-baseimage (used as root image for other components)
• https://quay.io/repository/mdeloche/power10-mma-inference-server
• https://quay.io/repository/mdeloche/power10-mma-stream-processing
• https://quay.io/repository/mdeloche/power10-mma-socketio-server
• https://quay.io/repository/mdeloche/power10-mma-mqtt-realtime-chart

The recommended way is to use these images and the run_demo.sh script. Instructions to run manually are also provided below.

[Containers] How to run

Requirements:

• podman

Start demo

Use the run_demo.sh script, providing the URL of the Kafka broker from which to read DayTrader data:

KAFKA_BROKER="kafka://10.20.30.40:9092" ./run_demo.sh

Stop demo

./stop_demo.sh

[Manual setup] How to run

Environment setup

• ensure that Git repository is cloned

cd AI_Component/

• install conda

wget https://repo.anaconda.com/archive/Anaconda3-2021.05-Linux-ppc64le.sh
sh Anaconda3-2021.05-Linux-ppc64le.sh -b
/root/anaconda3/bin/conda init && source /root/.bashrc

• restore "opence" conda environment

conda env create -f opence-conda-env.yml
echo "conda activate opence" >> /root/.bashrc && source /root/.bashrc

[BACKUP] With ausgsa channel:

conda create -n opence python=3.8
conda install flask numpy pandas scipy scikit-learn statsmodels kombu eventlet nodejs
pip install pytorch_forecasting faust python-socketio==4.0.0 python-engineio==3.2.0

1. Launch RabbitMQ (docker login docker.io might be required)

podman run -d --rm --name rabbitmq -p 5672:5672 -p 15672:15672 rabbitmq:3-management

2. Download trained model checkpoint using this link and place the n-beats.onnx file in the src/inference-server/ folder

3. Run inference server

cd src/inference-server/
PYTHONPATH=/root/AI-Usecases/src/training:$PYTHONPATH python3 inference_server_onnx.py

4. Run streaming service (update the KAFKA_BROKER variable to point to the URL and port of Kafka)

cd ../stream-processing/
KAFKA_BROKER="kafka://10.20.30.40:9092" python3 streaming_app.py worker -l info

5. Run SocketIO server

cd ../visualization/ && python3 ./socketio-server/server.py

6. Run Vue web server

git clone https://github.com/MaximeDeloche/mqtt-realtime-chart-client && cd mqtt-realtime-chart-client/
npm install && npm start

Note: the visualization range can be changed manually in the visualization/mqtt-realtime-chart-client/src/components/Home.vue file (min and max values lines 77 and 78).

6. Connect to http://server-ip:9080

Note: you might need to:

• disable SELinux: sudo setenforce 0
• open ports 9080 (web server) and 9081 (websocket connection): firewall-cmd --add-port=9080/tcp and firewall-cmd --add-port=9081/tcp

7. Launch JMeter. Connect to the IBMi system and run:

/QOpenSys/ocp5/jmeter/bin/jmeter -n -t /QOpenSys/ocp5/daytrader7.jmx -JHOST=127.0.0.1 -JPORT=10200 -JTHREADS=10 -JDURATION=120 -l jmeter_log.txt

Sources

• Nick Jokic - Real-time visualization of high-frequency streams