Pros/Cons of running *BSD

[Floid]

I have no idea what this mythical "straight" UNIX is (a Bell Labs tape on a PDP-11?), but if there's a kernel of truth to it, MySQL did/does 'assume' Linux, in the 'all the world's a VAX' sense, and there used to be some minor glitches there.

I'm not even sure if that applies anymore, and I've yet to become a database wonk myself.  People certainly run it, on all three BSDs.  The word from the horse's mouth might be a help.

As to selecting the 'right' BSD in this day and age... There should really be a 'product matrix' for this.  Hopefully I've got the following right:

NetBSD 2: BGL SMP, high-performance threading, fairly minimalist, "everything you could need" in pkgsrc, portable.  No linuxpluginwrapper, so use a Linux browser if you want Flash for desktop purposes.  Unsure as to 3D direct-rendering support, might exist for everything !nVidia.  Maybe a little shaky from the bump to 2.0, likely to keep doing whatever they've been doing as they always have.

OpenBSD: BGL SMP, fairly minimalist, secure, great for bastion hosts, firewalls, certain services.  Looks like performance for some workloads may be a concern, if you actually care that much.

FreeBSD 5: SMPng, high-performance threading, buzzword-compliant security (ACLs, TrustedBSD?), "everything you could need" in ports.  Still a bit warty from the whole 5.x adventure, may be continuing to improve in -STABLE.  Hanging out with DragonFly people gives you too many things to be afraid of.

DragonFly: LWKT, LWT someday?, libthr-like threading as of this week.  Infinite promise, while my current experience suggests you may wish to hold off 'production' use unless you know what you're getting into.  The next -RELEASE (1.1?) smells like it'll be a stable starting point when it arrives, so if you're not already a BSD nut, get on the bus then.  (...If you are already a BSD nut, drag out some spare hardware and help shake it down!)

Linux:  Looks good on paper.  There is probably some specific kernel point release that, coupled with some specific distro, does what you want.

I'm at a loss to tell people where to start right now; each and every distro *will* do the job, while the caveats present no clear winner.  (...and Linux sure *looks* like it's hit an island of stability right now, with Debian and Ubuntu seeming the moment's safe bets, but people who use it keep running to me with 2.6 horror stories...)

If you provide some sense of your mentality, astrological sign, and the lay of the lines on your palm, I might be able to peer into the crystal ball and make a guess as to which will be right for you.  Otherwise, there's always the argument for picking something and sticking with it until you find enough reasons not to... which is how everyone else gets along, whether they realize it or not.

[Floid]

Since I went through the trouble of digging up the MySQL results, here's the second part of that article with the pretty graphs.

I'm not surprised to see software that's mostly profiled on Linux working best on Linux; on top of this, FreeBSD 5.3 hasn't been very 'profiled' at all -- they've been too busy making the code work to check the theory in practice.  (This is a simplification, I'm sure individual developers are trying to pay attention, and I haven't been watching the development lists closely... but over the past couple years, Linux development has seemed much more 'benchmark-focused.'*)  The hope is that post-5.3, enough of the grunt work will be over with to actually take a look at what's working and smooth out the bumps.  NetBSD's SMP pessimism is a bit more interesting, and there are probably a couple variables at play there, too,

(Also note that, whatever the comments say, the 5.3 kernel does come built 'for 486' by default, not 386 -- which was said to impart some major performance concerns -- though that's a matter of kernel configuration, not CFLAGS.  I'd also wonder about the state of SMPng locking for particular disk and network devices used on the testbed, but hey, that's a real-world concern, too.)


*This can be good or bad -- it seems more scientific, but we know how much fun the GPU vendors have.  The 'problem' I've seen with Linux is that, every couple months, a new corner case comes into vogue, someone posts a bunch of benchmarks, a massive patchset appears to correct them, it gets committed, and suddenly 5 other corner cases break.  BSDs usually *try* to be more austere, and thus less of a moving target -- but all this business about FreeBSD 5 lacking a -STABLE until 5.3, and NetBSD murmuring 'oops, next time, let's open the patch branch immediately' means things haven't been all that much different lately.

[Floid]

Aha, NetBSD SMP results explained.