Well, simple: Consider, you have a multiplayer (online) game and want to find servers to connect to and to play on. You would not want to ask someone for the IP of a valid host. In most cases for online gaming you would not even get to know.
Any today's multiplayer game has that: Find servers, select one, connect. I guess, you know that, do you?
Marvin
Yep, I do know that. But this problem (and it gives way to several other issues) is outside the scope of what Mina handles. Mina is simply a low-level socket communication framework.
Since we definitely want to consider large scale, here's what I would assume is needed (in my limited experience):
Suppose you have a datacenter in West Coast US, East Cost US, and Central Europe. So the goal then is to give a list of servers to the user so he can pick one that he wants. Perhaps we want to visually indicate to the user which servers have the lowest latency, or which ones have the lowest population, or whatever, to help the user make a decision.
The client still needs to connect to a central server, let's call it Login Server. So as far as coding the client goes, it would simply connect to the Login Server, request a list of servers, and get a response. The host name and port must still be needed beforehand. This is the step in which I was referring to, and why I was getting confused
The task of monitoring a cluster's health, or which servers are down/available should all be done on the server side.
So then, we can assume that the Login Server would probably utilize a utility to make it easy to return a list of servers and the statistics we're interested in displaying back to the user. Something on the server side still needs to know about all of the servers. It's possible to make the servers a part of the same multi-cast group. Thus, when a server is brought up, it would join the group and broadcast a message (probably to a server controller of some sort). Of course, all of this is very dependent upon load balancers, if there are clusters, if there are servers delegated to certain zones, etc. When a user chooses a server, the Login Server or some other designated server on the hosting side would return the correct data center's IP. The client would then use the returned IP address for it's socket. This is just one option. The point is, this technology needs constructed outside the scope of Mina.
I know Darkstar addresses some of these issues, but I axed using that option for myself for a couple reasons. Arianne looks ok but I prefer the ability to create my own protocol and transport. I don't want to be forced to use UDP and their packet format (which should be sufficient for most).
General Category
