Developper Documentation
Targets
MCU Targets
Handsake Protocl

Handsake Protocl

This page describes a simple handshake protocol that can be used to communicate between the target and the host. The protocol is based on a simple request-response model.

Protocol

The suggested handshake protocol defines a simple way to exchange the input data, the metrics, warnings, and any runtime errors between the AI_Manager and MCU Target.

  • The target sends 0:START\n as soon as the target started.
  • The target sends 0:GET-DATA\n when it is ready to receive the input data that will be used to run the model.
  • The AI_Manager sends directly the right number of bytes
  • The target sends back the metrics in json format. The format of the message is the following : 1:{jyson_type:"inferance_metrics", ...}

Handling Errors, warnings and information

  • The target sends an info message like : 2:<YOUR INFO MESSAGE>
  • The target sends a warning message like : 3:<YOUR WARNING MESSAGE>
  • The target sends an error message like : 4:<YOUR ERROR MESSAGE>
⚠️
Important : With the suggested AI_Manager module, the errors messages 4:<YOUR ERROR MESSAGE> trigger a system error that stops the manager and the CI/CD. Unknown messages, (messages that doesn’t start by 0, 1, 2, 3 or 4) will trigger an error and stops the CI/CD.

Handling Hardfaults

In case of a hardfault, the target must be rebooted. The 0:START\n message is used to detect a Hardfault error. You must ensure that the target is able to send this message as soon as it starts (after initialization of the necessary resources). If the AI_Manager get this message after receiving the command 0:GET-DATA\n, then it means that the target crashed. And thus the benchmark should stop and an error should be thrown.

Message type

  • Command : message start with 0
  • Result : message start with 1
  • Info : message start with 2
  • Warning : message start with 3
  • Error : message start with 4
TargetsDocker