Pasted 7 months, 2 weeks ago — Expires in 141 days
Hello Allan/Benedict/JT, Thanks for the time and effort you put into the show. Matt Dillon of DragonflyBSD, not the actor, would be a very interesting interview. Or, if that's not possible, just a discussion of the merits and/or drawbacks of the design decisions in the Dfly kernel would also be very interesting. When Dfly was created, MD made some really strong technical arguments in support of his design for a multi-threaded kernel. From my understanding, essentially minimizing the use of the traditional fine-grained-locking scheme and instead using critical-sections whenever possible to achieve the same effect. If I recall correctly, one of MD's main points was that a "traditional design" multi-threaded kernel, although a proven scheme, was very complicated and that made it too difficult to optimally maintain by a project with a smaller group of kernel developers like FreeBSD. MD was sceptical on the ability of such a "smaller group" to maintain a kernel, with a complicated locking-scheme, in terms of scalability/performance on par with competing projects having more resources. I think his concern in that regard was when/if CPUs had what could be hundreds of cores. It's several years "down the road" now and it'd be interesting to get some detail on which of MD's original Dfly kernel design ideas have shown to have merit and which did not. It seems that the target use-case for Dfly is more toward clustering and maybe that's the niche where it makes such a design necessary and/or advantageous. It'd be good if the discussion didn't turn into a comparison of good vs. evil. There's always trade-offs. Thanks for the show.