madhuatberkeley / w251_homework3

Homework 3

On Jetson NX

1. Create a bridge network docker network create --driver bridge hw03

2. MQTT broker

   • Build the image

     sudo docker build -t nx_mqtt_broker -f dockerfile_mqtt_broker .

   • Spin up container and establish broker

     sudo docker run --rm --name nx_mqtt_broker --network hw03 -p 1883:1883 -ti mqtt_broker /usr/sbin/mosquitto

3. MQTT forwarder

   This spins up the forwarder that connects with both broker, so make sure the broker is up on VSI as well (see below).

   • Build the image

     sudo docker build -t nx_mqtt_forwarder -f dockerfile_mqtt_forwarder .

   • Spin up container and run nx_forwarder.py

     sudo docker run --rm --name nx_mqtt_forwarder --network hw03 -v ~/work/W251/v3/week03/hw/:/home/ -ti nx_mqtt_forwarder /bin/sh /home/nx_forwarder.sh

4. OpenCV face detector

   • Build the image

     sudo docker build -t nx_face_detector -f dockerfile_face_detector .

   • Spin up the container and run nx_face_detector.py

     sudo docker run --rm --privileged -e DISPLAY --name nx_face_detector --network hw03 -v ~/work/W251/v3/week03/hw/:/home/ -ti nx_face_detector /bin/bash /home/nx_face_detector.sh

   • Note that in nx_face_detector.py I added in a timeout of 5 seconds as well as a single digit counter to keep track of faces/pictures. These were used just so that I don't overwhelm the output with tons of images. Tests have been done with these two restrictions removed and it still works nicely.

On AWS

1. Create a bridge network sudo docker network create --driver bridge hw03

2. MQTT broker

   • Build the image

     sudo docker build -t cloud_mqtt_broker -f dockerfile_cloud_broker .

   • Spin up container and establish broker

     sudo docker run --rm --name cloud_mqtt_broker --network hw03 -p 1883:1883 -ti cloud_mqtt_broker /usr/sbin/mosquitto

3. MQTT receiver

   • Build the image

     Unfortunately, I didn't find a way to decode the bytes message without openCV, so this receiver image is the same with the image for face detector, except that I also needed to add s3cmd. It is bigger than I idealy wanted, but it works for now.

     sudo docker build -t cluod_prcessor -f dockerfile_cloud_processor .

   • Spin up container and run cloud_procssor.py

     sudo docker run --rm --name cloud_processor --network hw03 -v ~/W251/HW/hw03/:/home/ -ti cloud_processor /bin/bash /home/cloud_prcessor.sh

4. Note on S3 buckets

   The newer version of S3 buckets support public access much easier. s3cmd still works with the newer buckets, but one needs to create new credential with HMAC checked to see the access_key and secret_access_key.

Submission

1. The repo for the code can be found at https://github.com/MadhuAtBerkeley/W251/v3/tree/master/week03/hw, please let me know if there is any trouble accessing it.

2. The link to the faces can be found at

3. Naming of the MQTT topics: I created a simple single-level topic for the MQTT topic .

4. Choice of QoS: I picked QoS 0 for this task, which is also commonly known as "fire and forgot".