EDGESTREAM Dock

[Takenori Sato]

Hello developers,

EDGESTREAM Dock is another type of application that will be supported only on AI Box Advance.

It is a docker container that will be launched as follows.

$ sudo docker run --gpus all --rm --dns 8.8.8.8 --network STREAM_BRIDGE_NETWORK -v STREAM_FOLDER_ON_HOST:/app -v /tmp/.X11-unix:/tmp/.X11-unix -e DISPLAY=$DISPLAY -w /root -e location=RTSP_URL CONTAINER_NAME

Each meaning of those arguments are:

• --gpus all                                                         a Tesla T4 graphics card on an AI Box Advance box is accessible
  
• --dns 8.8.8.8                                                    a default DNS server is set
• --network STREAM_BRIDGE_NETWORK    a dedicated bridge network for this stream is created and set (the default bridge network allows interactions between containers, but which is not allowed here)
  
• -v STREAM_FOLDER_ON_HOST:/app         a dedicated folder on an encrypted folder of a host for this stream is passed as /app on a container
  
• -v /tmp/.X11-unix:/tmp/.X11-unix                   (only for testing) a container is allowed to access a X11 process on a host
• -e DISPLAY=$DISPLAY                                  (only for testing) a display variable on a host is passed as an environmental variable
  
• -w /root                                                            a working directory of a container is set at /root
  
• -e location=RTSP_URL                                  a rtsp url for this stream is passed as an environmental variable
  

A simple Dockerfile to illustrate how this works would look like this.

FROM nvcr.io/nvidia/deepstream:4.0.2-19.12-base
ENV location rtsp://127.0.0.1
CMD gst-launch-1.0 rtspsrc location=${location} ! rtph264depay ! queue ! h264parse ! nvv4l2decoder ! nvvideoconvert ! "video/x-raw,format=RGBA" ! nveglglessink

This is a Dockerfile that launches a simple GStreamer pipeline with a given environmental variable location.

Any data you want to keep after stop can be saved on /app.

Please note that a Tesla T4 graphics card has no display port like HDMI or Display Port.

So, you can not achieve various graphics operations in a container on an AI Box Advance.

Available services from the Edge AI Service to the EDGESTREAM Dock application are:

1. license management
2. installation/uninstallation operations

No other rich features like an encryption of AI model binary, Edge View, actions&events are not available.

But a container has no restriction on network access.

Please let us know if you have any feedback.

Thanks.

[ryosuke....@gmail.com]

Question. Any technical reason or policy to support Docker only with Advanced AI Box? Thanks.

[Takenori Sato]

No, there is none technically. 

So, we are open to your suggestion.

[Takenori Sato]

> I suppose to use standard AI Box if EDGEMATRIX supports docker with one. What are next steps as procedure? Thanks.

Thanks for your proposal. I will tell our contents dev team to evaluate.

2020年8月4日火曜日 11時13分16秒 UTC+9 Takenori Sato:

[ryosuke....@gmail.com]

Noted. Waiting for your feedback. Thanks.

2020年8月4日火曜日 13時25分48秒 UTC+9 Takenori Sato:

[Ryosuke Ishihama]

Hi,

Would you please let me know target date to evaluate Docker with Standard AI Box? With your information, I'm gonna make plan to procure either of Advance or Standard device.  

Thanks,

Ryosuke

2020年8月4日火曜日 13:25:48 UTC+9 ts...@edgematrix.com:

[ts...@edgematrix.com]

Hello, 

This is a very old post about EDGESTREAM Dock, which allows you to run any arbitrary code as long as it can run in a docker container.

Now we have a little more details. Please give us your feedback from developer's point's of view.

Service Model Comparison

At first, here's the service model comparison. We expect this EDGESTREAM Dock will be positioned somewhere between EDGEMATRIX Service and Azure IoT Edge/Greengrass. It would be more suitable for already established existing on-premise systems. 

EDGEMATRIX Doc Specifications Draft

2020年4月26日日曜日 7:38:24 UTC+9 ts...@edgematrix.com: