Resource optimization with DRL-driven real time service placement strategy in Edge-cloud continuum.

In a smart home environment, we are using cloud computing for powerful processing and storage of big data that are generated from IoT devices. Further we are also using Fog to handle the data in a more real time manner. Whenever we request for any service, the service provider take advantage of the resource available at fog and cloud. Fog resources are used to provide more real-time, sensitive services that require less computing and storage resources. On the other hand, cloud resources are used to keep the data (big data) for future purpose and process them. Sometimes, it is difficult to fulfil all SLA constraints only with the cloud resource and that's where the fog resource comes into the picture.

This thesis topic mainly focuses on efficient utilisation of cloud and fog resources. An AI model (most probably the DRL) needs to be developed to decide when to use fog computing and when to use cloud computing.

DRL-based Service Placement

Source Code

Experiment Applications

1) Application 1 - Deep learning based object classification using YOLOv3

Client Side
Server Side
Testing scripts
- curl -X POST -F image=@traffic.jpg 'http://$IP:$PORT/api/detect' --output traffic_yolo.png

2) Application 2 - Text-audio synchronisation or forced alignment (Aeneas)

Client Side
Server Side
Testing scripts
- python3 -m aeneas.diagnostics
- python3 setup.py build_ext --inplace
- python3 aeneas_check_setup.py
- curl -X POST -F audio=@p001.mp3 -F subtitle=@p001.xhtml 'http://$IP:$PORT/api/sync' --output p001.json

3) Application 3 - Speech-to-text conversion (PocketSphinx)

Client Side
Server Side
Testing scripts
- curl -X POST -F audio=@p001.mp3 'http://$IP:$PORT/api/sync' --output output.txt

Deployment Instructions

Clone project:

git clone git@github.com:chinmaya-dehury/AI4FogCloudServiceDisperse.git

Deployment of Environments

Goto nodes_apis
In Smart Gateway, run:

./execute.sh smartgateway.py

In Fog Master Node, run:

./execute.sh fog_master_node.py

In Fog Worker Node, run:

./execute.sh fog_worker_node.py

In Cloud, run:

./execute.sh cloud.py

Deployment of Application 1

Goto app1_yolo/app1_server
In Fog, run:

docker build -t app1-fog -f DockerfileFog .
docker run -d -p 0:5000 app1-fog:latest

In Cloud, run:

docker build -t app1-cloud -f DockerfileCloud .
docker run -d -p 0:5000 app1-cloud:latest

Deployment of Application 2

Goto app2_aeneas/app2_server
In Fog, run:

docker build -t app2-fog:latest .
docker run -d -p 0:5000 app2-fog:latest

In Cloud, run:

docker build -t app2-cloud:latest .
docker run -d -p 0:5000 app2-cloud:latest

Deployment of Application 3

Goto app3_pocketsphinx/app3_server
In Fog, run:

docker build -t app3-fog:latest .
docker run -d -p 0:5000 app3-fog:latest

In Cloud, run:

docker build -t app3-cloud:latest .
docker run -d -p 0:5000 app3-cloud:latest

Deployment of DRL Algorithm

Goto drl_algorithm
Update IPs/PORTs of both environments and applications in h_configs.py
In Smart Gateway, run:

./s_install.sh
./s_execute.sh