1. 21

  2. 4

    While I’m glad to see progress on this integration, I don’t see whether the hardest parts have been figured out. In particular, folklore is that IPFS lacks the bandwidth, and possibly the space, to host Nix’s entire public binary cache. Reading the grant proposal, those hard parts are tucked behind “stretch goals”. Has IPFS gained capacity recently?

    1. 5

      It depends on what is considered the hardest part I guess. Addressing all of the problems at once is the best way to achieve nothing. Even if this experiment doesn’t succeed, the work already brought a lot of Nix internals cleanup and will likely result in the addition of CA-derivations.

      IPFS itself doesn’t really store anything permanently. It’s missing content duplication and reward mechanisms that allow objects to outlive individual bridge failures. I think it’s what led to the creation of Filecoin. In order to publish the whole existing binary cache, the NixOS project would have to create a new kind of bridge that exposes all those files onto the IPFS network. Hopefully by lazily-fetching them from S3 when they get addressed.

      1. 3

        IPFS lacks the bandwidth, and possibly the space

        IPFS is a protocol. Bandwidth and space come from the hosts, just like with HTTP (and bandwidth can be temporarily added by everyone who fetches over the protocol, just like with BitTorrent).