"auto" server mode is inefficient
Currently the "auto" server mode is implemented as follows:
- Maybe act as server
- If there is a bus server object for the endpoint, use it
- If not, try to open a listen socket (for the configured endpoint) *
- Maybe act as client
- If there is a bus client object the end point, use it
- if not, try to connect to a listen socket (for the configured endpoint) *
The operations marked with "*" are potentially expensive (due to nameserver queries etc.).
When a process acts as client, steps 1.1, 1.2 and 2.1 are repeated for each created participant.