一个如何在AWS上扩展专用EC2实例？

Question

I'm creating on online real-time multiplayer mobile game using Kryonet (a Java TCP/UDP networking library) that I'm planning to host on AWS.

The architecture is as follows: clients connect to a central login/account server that allows them to login and view their stats etc. This bit is easy, as it'll basically just be a REST API, and can be scaled in a pretty standard way (like you would any webapp).

However, the more interesting bit is when players actually play a match. For this, I plan to have a separate pool of "match" servers (EC2s). The login/account server will pair two players, then send the client the address of a particular match server. The players will then join that match server, which will host their match (perhaps lasting 5-10 minutes). The match server needs to be sticky as it will be running a real-time instance of the game, and will be sending/receiving UDP packets in real time. Each match server will probably be able to host a few hundred matches.

My question is about how I should go about scaling these match servers. I suppose I will have them auto-register with the central server at start-up, and send some type of keep alive. I could build this all myself; however, I'm wondering if AWS has tools/services that can do this all for me.

Answer 1

Okay, I've done a little more reading of the AWS documentation. It seems that I can probably achieve this as follows:

Each time two players are paired up, they are added to a queue. They are taken off the queue when a spot is free on one of the match-playing servers. When the size of this queue exceeds some threshold, then the number of EC2s is scaled. This can be done with basically all in AWS config: http://docs.aws.amazon.com/autoscaling/latest/userguide/as-using-sqs-queue.html

The tricky bit is then scaling down instances. Unlike a normal REST API, you can't just turn a server off. The server needs to finish all its current games. It seems that AWS has this covered too with lifecycle hooks.