Matthew Dillon posted an extensive writeup about the hardware changes for dragonflybsd.org; price to performance ratio has been improving so much for multiprocessor machines that we can jump forward both for hosting hardware and for a testbed.
He also mentions his immediate thoughts on what to tackle next, since SMP has been so relentless improved in DragonFly. It resulted in a very long conversational chain as people weighed in with opinions, so I’ve held off posting it until the conversation finished. (I chimed in too.)
The jabs at OpenBSD and NetBSD are spectacular.
I’m waiting for the fight to ensue. I’ve got my popcorn out ready to watch.
Jabs? He mentions he doesn’t know what their SMP plans are, and that’s it. I’d describe that more as “a question”.
I wonder what you’ll link when there won’t be mailing lists, gitweb? Will dillon be a returned daniel robbins posting on a forum? I don’t even visit the funtoo forums anymore, felt weird. And isn’t there some forumlike plugin for ikiwiki?
Re: SMP
With Dillon claiming lockless have now been eliminated and SMP is a solved problem … it seems only appropriate that we get an updated perf benchmark of Dfly vs Linux.
Benchmark it yourself, cause asking repeatedly for someone, somewhere to do it for you will not work.
I can’t tell if you are asking that repeatedly because you think it’ll work or because you’re trying to turn it into a joke/meme.
Anonymous isn’t alone in wanting updates Postgres benchmarks. I do too.
That’s fine, but you are asking for it in a way that’s guaranteed to not change anything. I can’t do it; I don’t have time. It’s very frustrating to me to read people asking for something I can’t supply, and then doing it over and over again.
The benchmark methodology was documented when Francois did it, and it doesn’t require anything beyond some spare hardware and time, so it feels like you and others are being lazy and asking for a fix for a problem you can fix yourselves, much faster, and then purposely asking wrong on top of that.
The writeup briefly mentions synth. I wonder though, what point there is behind the recent news of a synth rewrite in C: https://www.phoronix.com/scan.php?page=news_item&px=DragonFlyBSD-DSynth-Builder
Is anything known about technical reasons for such a rewrite in an arguably inferior implementation language, that cannot be handled in the Ada implementation?
dflybsd builds from source, dflybsd is written in C, dsynth will be used to build dflybsd, base and ports. inferior? you can still use synth or ravenports, and like ravenports and unlike synth, dsynth will have a web-based dynamic build report.
the greatest benefit of dsynth will be dillon’s. what? speed. why? because he talks about upgrading a computer that synths in a week to a computer that synths in over a day. a day! dsynth would make that computer finish in under a day!
From superficially looking at it, just editing the Ada implementation to achieve the C rewrite’s supposed benefits still looks like a more reasonable way to get to the desired outcome.
I don’t think it’s a modify it to work and forget kind of thing, dsynth will be developed futher, it is tightly coupled to the most modern smpnuma bsd, dflybsd, but the plumbing of it to play above well with everything else is the hardest. dillon is constantly messing with dsynth right now on the dragonflybsd’s git and it is probably not even halfway done, who would do so much work in ada? dfly’s repository has mostly C code. if people start doubting the tools written by dillon, it means the world just turned upside down..
Can’t wait to have a functional graphic driver for
AMD A8-7650K Radeon R7 4C+6G (4) @ 3.300GHz
Is the only thing that keeps me kissing Linux goodbye from my workstation.