RPC performance in large network

I have a large network, supposed to manage ar开发者_Python百科ound 8000 nodes (the requirement is for client server based filesystem) . Which of the following implementation is better - "RPC (ONC RPC) using TCP" or "direct socket using TCP"?

Comparison points should be:

1. memory

2. bandwith issues

3. Performance

4. Any disadvantage while using RPC

In general, there is a little more memory, bandwidth, and performance overhead when using RPC. However, RPC can greatly simplify your logic.