DragonFly version 6.4 is the next step in the 6.x release series. This version has hardware support for type-2 hypervisors with NVMM, an amdgpu driver, the experimental ability to remote-mount HAMMER2 volumes, and many other changes.
The details of all commits between the 6.2 and 6.4 releases are available in the associated commit messages for 6.4.0.
The downloads are ready.
DragonFly seems pretty neat, I’ve always enjoyed the theory of it. But, I have no idea of any large usages of it. Can’t really convince anyone to give it a try in production. It might be a bit faster for some workloads, but orchestration and security updates probably aren’t the best right now.
I’ve never used it personally, but the HAMMER FS is something I’d be interested in giving a go. It seem like DragonFly put a lot of work into very high scalability, which is cool but like you I also find it hard to think of applications on current hardware. It seems like they could support significantly more hardware cores than we have today, which is neat, but I don’t know about real world utility.
In theory it opens up application development to massive SMP architectures, like having a process for every planet in a galaxy. I guess maybe if you had a game universe where every user could have their own dedicated process, it offers both scalability and more robust process isolation. But current hardware isn’t even close to scaling this large without reverting to clusters even if the OS can support it.
Yeah. Its really hard to for any start up to justify a on prem setup that would have epyc processors that would really make use of the SMP advantages. AWS you’d be better off designing it to not need a single beefy server.
Besides, for on prem, I’d really really like to give Oxide a try. it seems crazy, with a big upfront investment, but its a zagging when everyone else is zigging. Its like a self hosted cloud. But I’d trust Bryan Cantrill.