Mutual Blocking in C++ Multi-threaded Inference with CompiledModel Instances

[EunhaTaker]

When c++ multiple threads use a CompiledModel singleton, or when using multiple instances to infer, they both encounter mutual blocking issues. Can this be avoided through configuration? Would improper model compilation lead to this situation?

[ilya-lavrenov]

Do you use sync request.infer() call on multiple requests of the same ov;:CompiledModel?
If yes, prefer request.start_async

[EunhaTaker]

Thank you, I'll give it a try when I have some time.