OpenTafl v0.3.x: The Networkening Begins

The time has finally come. OpenTafl v0.3.x is officially under way. You can find the ongoing changes to the source code at the Mercurial repository. I’m going to spend this post talking a little bit about the structure and organization of the OpenTafl server.

First: I decided that a thicker server was the way to go. Peer-to-peer networking is almost always more trouble than it’s worth for things like this, and OpenTafl shouldn’t be all that resource-hungry. So, the OpenTafl server will maintain the authoritative version of each game, and the authoritative version of that game’s clock. OpenTafl uses TCP instead of UDP; OpenTafl doesn’t run a lot of network traffic or need particularly low latency, and the guarantees provided by TCP are super-handy.

Second: I wanted a server which could easily be scaled up and down. Rather than work with Java’s default threading model for networking, which uses a reader thread per client to both read and do the work, I decided to write a priority task queue. Everything the OpenTafl server does, from parsing incoming communication from clients to updating the game clocks to distributing chat messages, is encapsulated in an individual task object. The task objects are pushed to one of three queues, corresponding to high, standard, or low priority. Higher-priority tasks are executed preferentially, unless there’s a large backlog of lower-priority tasks. This lets the number of worker threads be subject to user configuration. Finally, to distribute load, the OpenTafl server spreads out repeated tasks over a number of buckets in the period of repetition: clients get an updated game list every 30 seconds, by default, but only one in thirty clients gets the game list per second.

Third: Clients are expected to do some heavy lifting. The client gets a server clock update every five seconds, and has to maintain its own clock for display purposes in between. Although the server runs the authoritative version of the game, each client also keeps a full copy of the game. This lightens the burden on the server, which does a lot, and pushes it to the clients, who don’t do much, relatively speaking.

Those are the big features. I’ve run some successful tests over the past day or two, successfully playing out games over the network. I have more features to do before I can make an initial public release, but I expect to have one for you in the next few weeks.

Leave a Reply