Inefficient RPC for unique connection/channel applications

[chrishenx]

There are applications where an unique connection and channel are used to communicate with a RabbitMQ server.

I measured ~500 ms when creating a connection, ~100 ms creating the channel. Clients could cache the connection and avoid the greater overhead. But still there's a quite large ~100 ms overhead.

It would be really nice if the first parameter of the rpc function were either a connection or a channel. If it is a channel then we probably shouldn't close it at the end of the operation.

What do you think?