Project Darkstar

» see projectdarkstar.com as it lived, on the wayback machine

~ the last words ~

Oracle closed its 7.4 billion dollar acquisition of Sun Microsystems on 27 January 2010. Within a week, a Sun representative posted to the Project Darkstar community forum that the project was over. The wording was the kind a company uses when the decision came from somewhere above the person typing: they would keep the source repositories and projectdarkstar.com alive "for as long as we can", and were "investigating other homes" for the code.

Darkstar had started in 1999 as Jeff Kesselman's personal project, then became a research effort inside Sun Labs. The first public release came on 30 August 2007. It was open source, written in Java, and sold as middleware for anyone building a massively multiplayer game or a virtual world. The named enemies in its own documentation were zone overloading and data corruption, which between them describe most of what went wrong in online games during that decade.

The last release was 0.9.11, dated 29 October 2009. Three months later the acquisition closed and the project did not survive the handover. Project Wonderland, Sun's virtual world toolkit, got the same treatment in the same clearout, though its team lead said publicly that they had seen it coming and had already been looking for somewhere to land.

~ the record ~

~ what it actually solved ~

The interesting part of Darkstar was its concurrency model. You wrote your game logic as tasks, in what looked like plain single-threaded Java, and the stack underneath ran everything in parallel and handled the transactions for you. Every task was persisted when it ran, so if a server died mid-flight the work could be replayed on restart in the state it had been in before the crash. Updating a character's inventory was not supposed to be able to corrupt it, no matter where the process fell over.

That is a hard problem, and the reason it is hard is that the abstraction has to hold under load or it is worse than useless. Darkstar never got the chance to be proven wrong at scale, because no commercially significant game shipped on it before Oracle turned off the lights. The version number was still 0.9.

~ the fork ~

The community took the last codebase and relaunched it as RedDwarf Server, inheriting Darkstar's licensing and its mailing list culture, which is what an open source project is supposed to be able to do when its corporate sponsor walks away. It worked for a while. Today projectdarkstar.com does not resolve, and neither does reddwarfserver.org. A SourceForge page for the RedDwarf download is the last public surface either name has.

~ the engineering it was aiming at ~

I write code now for machines that have to keep a coherent picture of the world while flying, with no data center behind them and no second chance on a dropped frame. Darkstar picked the same shape of problem from the other end: let the person writing the application pretend the world is sequential, and make the runtime responsible for the fact that it is not. I never ran a Darkstar server and was never going to, since I have never shipped a game. What is recognizable from here is the bet, which is that the right place to put the hardest code is underneath, in a layer nobody using it has to think about. Sun kept making that bet. Oracle bought the company that made it and spent its first week deciding which of those bets to stop paying for.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »