cross-posted from: https://lemmy.world/post/4930979

Bcachefs making progress towards getting included in the kernel. My dream of having a Linux native RAID5 capable filesystem is getting closer to reality.

  • @qprimed@lemmy.ml
    link
    fedilink
    English
    13
    edit-2
    1 year ago

    a striped hot ssd cache for a massive raided backend on a VM host - all done with native filesystem support sounds pretty good to me.

    • JWBananas
      link
      fedilink
      English
      2
      edit-2
      1 year ago

      You’re running local storage for a VM host? Or are you talking more like whiteboxing your own NAS?

      I understand what bcachefs does. I’ve used bcache many years ago to do exactly what you’re describing, albeit for bare metal servers. I’m asking why.

      I’m just trying to understand what the use case would be in 2023 outside of a home lab, given that cost per gigabyte is basically at parity between SSDs and HDDs when you consider TCO (i.e. when you price in the extra power and cooling overhead for the HDDs, failure rates, and such).

      • @qprimed@lemmy.ml
        link
        fedilink
        English
        3
        edit-2
        1 year ago

        I have a pretty strong usecase for distributed Small/Medium Business bare metal VM hosts. most locations do not need NAS/SAN, and DAS will more than suffice. lower cost hardware with a near-line raided backend and SSD hotcache at FS level seems to be a pretty decent sweet spot.

        obviously this is not some enterprise grade setup and YMMV, but I am pretty interested in a all-in-one FS solution. I am sure others may have more innovative setups where its even more interesting.

        • JWBananas
          link
          fedilink
          English
          31 year ago

          Sounds like a great use case.

          I’m assuming your nearline drives speak SAS? Are you doing redundant controllers on the backplane for multipathing and for fault tolerance? I’m not sure if bcachefs specifically supports it (or if that would happen at a different layer) but distros in general should support it.

          TCO gets split into CAPEX and OPEX, so you come out ahead on the initial purchase even though it uses more power in the long run, which surely looks better to the business.