Enclage Device Blueprint key demo

This demo performs a confidential machine learning inference on an edge device, by being This project is a part of a Blueprint for a Confidential Computing at the Edge.

The main function of the application is to perform an image classification between key images (Scalys, Arm and Microsoft). It can work both in a single shot mode clasifying an image file and in a video stream capture mode.

The significance of this demo is a protection of a sensitive payload (in this case - ML model) using Confidential Computing Blueprint.

Hardware platform

Demo setup consists of a TrustBox Edge 201 and an optional usb camera for trying out streamed operation mode. Though it also works on the other devices from the TrustBox Edge line.

Demo is built and tested with web camera Logitech C270. Potentially any generic USB camera should work, but reliable functioning is only tested on C270. Other cameras may experience worse classification quality, as the machine learning model was trained on the images captured from C270.

Build

Build is verified and tested on an Ubuntu 20.04 host system. For reprodution SDK with toolchain should be installed.

Build environment

To have a reproducible environment docker build environment from this project can be used. To prepare and run it perform:

$ cd docker
$ make

To run a session witihn this environment:

$ docker/run-20.04.sh

Toolchain

A successful build of this project requires a specific SDK to be used. Download it from Scalys releases here and install it via command:

$ chmod +x ./poky-glibc-x86_64-scalys-openenclave-image-aarch64-trustsom-toolchain-ext-3.1.10.sh
$ MACHINE=trustsom ./poky-glibc-x86_64-scalys-openenclave-image-aarch64-trustsom-toolchain-ext-3.1.10.sh

Application bulid

Source the toolchain:

$ source ~/poky_sdk/environment-setup-aarch64-poky-linux

Create a build directory next to the downloaded oeml demo:

mkdir edb-key-demo
cd edb-key-demo

Configure and build the application:

$ cmake ../edb-key-demo \
  -DPKG_CONFIG_SYSROOT_DIR="${PKG_CONFIG_SYSROOT_DIR}" \
  -DOE_TEE=OP-TEE \
  -DOE_PACKAGE_OPTEE_PLATFORM=trustsom \
  -DOE_OPTEE_PLATFORM=trustsom \
  -DCMAKE_TOOLCHAIN_FILE=../oeml/cmake/arm-cross.cmake\
  -DCMAKE_BUILD_TYPE=release \
  -DOE_PACKAGE_PREFIX=${PKG_CONFIG_SYSROOT_DIR}/opt/oe
$ make

The resulting build outputs are located at:

out/bin/1f574668-6c89-41b5-b313-4b2d85d63c9d.ta
out/bin/edb_key_demo

Run

To run the demo manually on a device in a single shot mode:

# ./edb_key_demo 1f574668-6c89-41b5-b313-4b2d85d63c9d.ta r.png

A successful execution should print

[ 0.00 0.00 1.00 0.00 ] -> CLASS RED: 1.000000

To run in a streamed video mode from camera /dev/video0 skip the argument:

# ./edb_key_demo 1f574668-6c89-41b5-b313-4b2d85d63c9d.ta

Successfull execution will produce a continuous stream of label names classified from the camera.

For more details see ./edb_key_demo --help.

Cloud management

For a one-button Azure cloud infrastructure deployment see this page.

Confidential Package Management

For information on installing and using Confidential Packages see this page.

Enclave Device Blueprint reference implementation

For instructions on preparing the TrustBox Edge 201 for a demo refer to the setup guide

For a full end-to-end Enclave Device Blueprint description, refer to the EDB demo guide