Some of the planned blackouts will be temporary, others plan to shut their subreddits down indefinitely in protest.

  • @argv_minus_one@beehaw.org
    link
    fedilink
    29
    edit-2
    1 year ago

    Problem: millions of redditors currently use third-party Reddit apps. Abruptly sending millions of people to the Lemmy instance you just deployed is a sure-fire way to break it, and maybe bring down the whole federated network of Lemmy instances. Lemmy currently has issues scaling above a few hundred users, as Beehaw has recently discovered, let alone millions.

    Problem: Lemmy is a completely different protocol, and there’s less than a month left before all third-party Reddit apps become useless and everyone uninstalls them. That’s an exceedingly tight timetable and an exceedingly unforgiving deadline.

    That said, it’s now or never; death or glory. We’re not going to get another chance to bring over that many people to Lemmy all at once.

    • 🦊 OneFluffyBoi 🦊
      link
      fedilink
      101 year ago

      @argv_minus_one Is that scaling problem a software issue, or a hosting issue? There are other Fediverse platforms like Akkoma that use Elixir, so maybe they’d fair better? Could also pick several federated instances to distribute users to.

      • @argv_minus_one@beehaw.org
        link
        fedilink
        181 year ago

        Lemmy is written in async Rust. The language isn’t going to create a scaling problem. Well-written async Rust applications have handled vastly heavier workloads than Lemmy without a hitch.

        There are, however, some serious performance bottlenecks that need to be dealt with, and it remains to be seen whether any more bottlenecks remain undiscovered in either the protocol or the implementation. To be honest, as someone working on a Rust+Postgres application myself, this is the sort of thing that keeps me up at night.

        Hosting can of course be an issue as well. I’m under the impression that Beehaw had to go up several tiers in its hosting plan in the last few days in response to the surge in demand. I assume this was done to work around the aforementioned bottlenecks by simply throwing more hardware at the problem, but I don’t know.