Choosing the optimal network communication protocol for Federated Learning with TFF

Question

I am working on a federated learning project for my university, and my goal is to implement it using TensorFlow Federated (TFF) to enable collaboration among multiple clients. However, TFF currently only supports simulation on a single machine, and I need to establish network communication capabilities from scratch to enable communication between multiple clients.

Considering this requirement, I am faced with the decision of selecting the preferred network communication protocol for my project. Should I use RPC (specifically gRPC) or REST? I would appreciate insights into the pros and cons of each protocol in the context of my problem.

Specifically, I would like to know:

1. What are the advantages and disadvantages of using gRPC for network communication in a federated learning setting with TFF?
2. What are the advantages and disadvantages of using REST for network communication in a federated learning setting with TFF?
3. Which protocol would be better suited for enabling communication between multiple clients in a federated learning setup, and why?
4. Are there any specific considerations or limitations related to the integration of either protocol with TFF and federated learning?

Thank you in advance!