I'll admit, I've only read a few passages, but I think I would have felt more comforted if they just said "We know and love illumos, and for many of these close calls that's enough." Because this RFD can sometimes feel like Linus's half-bigoted comments re: ZFS.[0]
For example, the systemd section especially could have been stolen from a Reddit thread of 5 years ago. I'm also willing to bet, simply because of its breadth of use, systemd is faster, as, if not more, stable, and much better documented than SMF. Linux's procfs and sysfs are similarly pretty great these days, and are also slagged nonetheless.
But my point was -- familiarity isn't a bad thing when you're building a cloud computer, and you've already built a cloud on illumos. Deep knowledge of dtrace should make you choose it over eBPF, and deep knowledge of FMA/SMF should make you choose them over systemd, especially if you're not confident in the alternatives.
Notwithstanding, since I'm not buying a $1 million rack, I'm actually much more interested what, if any, portable OSS tooling will be built around this computer. Because IMHO if anything was wrong re: Joyent, it was that not much of its OSS technology saw wider purchase, and one reason was -- it was and is exclusive to illumos. My (backseat QB) fear is Oxide may have fallen into the same trap. As mobilizing as a Cantrill-ian presentation may be, (again I'd bet) `ripgrep` created thousands upon thousands more Rust programmers and 100x the mindshare. illumos may be a perfect fit for this rack scale computer, but can you imagine the mindshare gained by simply having all your cool new tooling work on Linux? As we know re: an exciting new company, mindshare means a lot, so ... `statemaps` for everyone?
> I'm actually much more interested what, if any, portable OSS tooling will be built around this computer.
The interface to the rack is built around the classic API-first strategy. We provide an OpenAPI document describing the whole API, and the console and clients are all generated from it. HTTP is, of course, cross platform.
> illumos may be a perfect fit for this rack scale computer, but can you imagine the mindshare gained by simply having all your cool new tooling work on Linux?
illumos is not a user-facing aspect of the rack. There's no reason that tooling would be illumos specific. The API clients I refer to build on Mac, Windows, and Linux (the CI runner is Ubuntu but have no reason to only support it) as well as illumos.
For example, the systemd section especially could have been stolen from a Reddit thread of 5 years ago. I'm also willing to bet, simply because of its breadth of use, systemd is faster, as, if not more, stable, and much better documented than SMF. Linux's procfs and sysfs are similarly pretty great these days, and are also slagged nonetheless.
But my point was -- familiarity isn't a bad thing when you're building a cloud computer, and you've already built a cloud on illumos. Deep knowledge of dtrace should make you choose it over eBPF, and deep knowledge of FMA/SMF should make you choose them over systemd, especially if you're not confident in the alternatives.
Notwithstanding, since I'm not buying a $1 million rack, I'm actually much more interested what, if any, portable OSS tooling will be built around this computer. Because IMHO if anything was wrong re: Joyent, it was that not much of its OSS technology saw wider purchase, and one reason was -- it was and is exclusive to illumos. My (backseat QB) fear is Oxide may have fallen into the same trap. As mobilizing as a Cantrill-ian presentation may be, (again I'd bet) `ripgrep` created thousands upon thousands more Rust programmers and 100x the mindshare. illumos may be a perfect fit for this rack scale computer, but can you imagine the mindshare gained by simply having all your cool new tooling work on Linux? As we know re: an exciting new company, mindshare means a lot, so ... `statemaps` for everyone?
[0]: https://news.ycombinator.com/item?id=22005181