I guess it is to let the client in the primary server connect to the server in the two client and poll data from the subordinate server, 
right?
That is a considerable one
@particle_adam, polling allows you more control over what and when you get data from your nodes. The nodes simply gather data which the primary Photon calls for at fixed, round-robin intervals. This way you can use TCP and decide on the polling interval.
The other way is to use multicast UDP where ALL nodes are listening. The primary Photon can multicast a request that contains a specific node ID so that only THAT node responds to the multicast. Again, the primary Photon will do round-robin polling. The biggest difference is that UDP is a “lighter” protocol, requiring you to do any error checking in your code. On errors, you could simply drop the returned data until that node is polled again. If you can’t tolerate dropping data, then you will need to query the node again at the cost of extra time in your polling sequence.
Hope that makes sense! 