This is one of the basic problems in distributed computing, and there are two approaches that work, to a degree:
Registry Service
Somewhere on the network, you run a Registry Service with a well-known host and port number. This service has to be reachable / addressable from every place you want to run the application.
On startup each instance of the application on the network registers itself with the registry.
When some machine / program needs to locate an instance of the application, it asks the registry.
Problems:
You have to deal with application instances that "go away" without telling the registry.
You have to have a way to restore state if the registry restarts.
The applications have to know the name (or address) and port of the registry instance.
Broadcast / Multicast
Every instance of the application listens on a well-known "broadcast" or "multicast" address / port.
When a program wants to locate instances of the application, it sends a broadcast / multicast request.
Each instance responds to the request giving its details.
The program accumulates the responses to build a list of all "live" instances.
Problems:
This doesn't scale. Each and every request from M programs goes to N machines and generates N responses. As M and N grow, the network traffic grows quadratically.
Broadcast and Multicast are lossy, especially on busy networks.
Broadcast typically doesn't cross network boundaries. Multicast requires special configuration.
Either approach should work on a small network with a limited number of instances.
The simple approach is to identify an existing distributed computing technology that does most of the work for you. For example, RMI and a RMI registry, dynamic DNS, CORBA, JINI.
与恶龙缠斗过久,自身亦成为恶龙；凝视深渊过久,深渊将回以凝视…
