NetBSD-Current Gets SMP

MobyTurbo writes "NetBSD-current for the i386 architecture now has SMP. (It used to be that only FreeBSD had this feature among the free BSDs.) See the announcement on the current-users mailing list."

Comments

x86 only?

[GOD_ALMIGHTY]

When will the Dreamcast Port get it?

I really want to try this out on my quad-proc Dreamcast.

</sarcasm>

Re:x86 only?

[Dave9876]

Nah, if only they'd get it working on the sparc port, then I'd have a reason to go out and find another SM51 processor for my SS10...

This can only be a good thing

[Tim_F]

Finally OpenBSD will ahve some straight up competition. For a long time it has been the most secure, and the only BSD with SMP support. Now we will finally see the battle for BSD supremacy heating up!

Can't wait to see what FreeBSD does to top this!

Re:This can only be a good thing

[jonestor]

According the the summary at the top, it's FreeBSD that has it and OpenBSD doesn't.

Re:This can only be a good thing

[MobyTurbo]

Blockquoth the poster:

Finally OpenBSD will ahve some straight up competition. For a long time it has been the most secure, and the only BSD with SMP support.

Can't wait to see what FreeBSD does to top this!

According to the official OpenBSD FAQ, OpenBSD does not have SMP. Either in -CURRENT (development branch) or in release form, though apparantly there is a group working on it. FreeBSD on the other hand will have an improved fine-grain implementation of SMP, in their upcoming 5.0 release, and already have a more primative version in the 4.x releases. It's really the reverse, OpenBSD is the only free *BSD *without* SMP being tested. I have no idea why you thought otherwise.

Re:This can only be a good thing

[dohcvtec]

OpenBSD has never had SMP, and if you read any of the OpenBSD mailing lists, you will see that it is not a priority to have SMP. And there is no battle for BSD supremacy; the BSDs peacefully coexist, each standing on its own merits.

yawn

[honold]

call me crazy, but i could really care less about smp. i would wager the wide majority of smp systems fall into 2 categories:

1) unnecessarily powerful servers
2) unnecessarily powerful home braggart systems

database servers? sure. heavily loaded web servers? sure. file servers? NO. desktops? NO.

at least the scsi bigots will actually net some measurable performance increases if they drop some money on a 15k drive.

i sincerely hope openbsd continues to focus on OTHER things like openssh - you know, that thing you probably use every day of your life on your non-smp machine?. since most openbsd boxes are used as edge devices, the only big need for processing horespower is in crypto...

and that problem can be solved by purchasing a hifn-based pci crypto accellerator for $90 from soekris.com, thanks to openbsd's excellent hardware crypto accelerator support.

once you get past the crotch-grabbing aspect, low-end smp is not what most of the world would have you believe it is. high-end smp will likely get replaced by clustering of commodity hardware.

Re:yawn

[OrangeSpyderMan]

call me crazy

Ok, crazy guy. :-)

I'll bite. I am currently running two SMP machines. Ok, one is for a DB server. Leave that one out of the question, it is actually underpowered for much of what it does, alas. I'm saving my pennies. The other however is my main desktop. It is not unecessarily powerful (2xPIII 600EB) compared to todays desktops systems, on the contrary, but I would never swap it for a single 1.2GHz (or even higher - a 2.8 P4 might get me thinking :-) ).

The reason is simple. On my desktop I frequently have a number of concurrent processes running (Mozilla, compile, ogg player, a few ssh sessions and I might even fire up a game from time to time while I'm waiting for a compile to finish...). This kind of use shows what a boost "low-end" SMP can be - the system remains perfectly responsive way past loads that would have a similar "horse-power" single CPU system groaning - and that is very important for interactive desktop use. My box at the office is a single PIII 1 GHz, which should, on paper, hold its own quite well. It feels markedly more sluggish for desktop use.

SMP systems are little more prone to "pissing competition" type purchases than say, GeForces and P4. I don't know many people who can actually use all the horsepower of modern systems on the desktop, be it under *BSD or Linux. As someone once said, todays desktops just "wait faster". At the moment at least I'd take a lesser CPU 2-way SMP system over a more powerful single CPU for my desktop anyday.

Re:yawn

[Foresto]

While I agree that most home systems don't have much use for an SMP system, you are making an overly broad generalization. When you say "desktops", you're including a whole lot of computers and their users, many of whom do more with their systems than browse the web and play games. Compiling software is one task that my computers regularly perform, for both work and personal projects, and let me tell you, SMP can greatly reduce the number of hours per week I spend waiting for my computer to finish. Video encoding is another task that some of my friends do at both home and work, and it greatly benefits from a second CPU. I'm sure there are other such tasks that, while not mainstream activities that mom and dad would be doing in the living room, are certainly common.

Port to OpenBSD soon perhaps?

[moonboy]

I wonder if this means OpenBSD will soon have SMP capability? Anyone have any thoughts? Inside information?

Re:Port to OpenBSD soon perhaps?

[Dokta_C]

Hell, I'd be happy if they could get it into NetBSD/sparc.

Re:Port to OpenBSD soon perhaps?

[LizardKing]

NetBSD/sparc has already supported SMP for ages.

Chris

Re:Port to OpenBSD soon perhaps?

[Dokta_C]

I thought it was only spinup?

Re:bad news for Linux?

[glenstar]

(While I assume that the latter part of the above message is tongue in cheek...)

I guess it depends on what sort of enterprise application you are talking about. I am building a commercial router/content filtering system for corporations using NetBSD and am *very* happy with it.

NetBSD is small, robust, and fast. Creating customized installation media is a breeze and the networking code is fast (okay, I haven't benchmarked it against any of the 2.4 Linux kernels, but it *seems* faster). Hell, my full ISO to install NetBSD and all of the other supporting software is approximately 100MB, and that's with quite a bit of extra stuff thrown in, like the development tools, etc...

The one place that NetBSD needs some help is with the installation process ("you mean I need to know the geometry of my drive???!" and "what do you mean sshd isn't started by default from rc.conf???") but even that is easily overcome by creating simplified installation floppies.

As a recent convert to NetBSD (from FreeBSD from Linux), I have to say that I am very pleased with the NetBSD product... adding support for SMP can only make it better.

Re:bad news for Linux?

[rplacd]

Actually, many embedded system developers use NetBSD for their eval boards and such.

Also, at least one major company has used NetBSD in the past -- DEC, for their DNARD ("shark") systems.

netbsd and smp

[rplacd]

Note that while the i386 port just got SMP support, other ports have had it for a while. NetBSD/macppc got it in August, NetBSD/sparc got it over a year ago, etc.

Wow!

[Dimwit]

Now BSD can die twice as fast!

(Note to moderators: Not a troll - just an sad attempt at humour. I'm writing this from my FreeBSD 4.6.2-RELEASE box...)

Silly BSD question / Mac question

[lonely]

At work I have a dual processor 'Windtunnel' Apple Mac running Mac OS X. Does this not count as SMP? Or have I misunderstood the term?

(Dawin is of course a flavour of BSD)

Re:Silly BSD question / Mac question

[josepha48]

Hmm well sort of.. it actually is build on the mach kernel which was a fork of bsd way back when.

Of course this is NET-BSD not BSD. FreeBSD I believe has had SMP support for a while now. FreeBSD is more like Darwin / Mac OS X than NetBSD. Also they are referring to the i386 which is way different from Mac which uses the motorola processor.

NetBSD runs on just about all processors out there, does Mac OS X? No and neither does FreeBSD. That is what the whole NetBSD project is about. Mac OS X is about a pretty gui on the foundations of UNIX / BSD. Kinda about time someone did what Mac did, but then again about a year before Apple announced their plans of OS X I suggested that someone put a nice GUI on UNIX. Guess what. They listened and now everyone is really taking to Mac.

Re:Silly BSD question / Mac question

[castlan]

When people on slashdot speak of BSD, they tend to mean the post 386BSD derivatives. That was the "Free" OS that sprung NetBSD and FreeBSD. Then OpenBSD branched from NetBSD. Until fairly recently, none of these were SMP capable. For a while, only FreeBSD was SMP capable, now it looks like the "Free" BSDs (386BSD derived) are all rounding out, and starting to overlap in scope. FreeBSD now runs on non x86 hardware, and NetBSD now runs SMP. OpenBSD has various groups bringing SMP (mostly from NetBSD work) and NetBSD/FreeBSD are reincorporating OpenBSD code-audit correction. On the whole, it is a very pleasing situation.

Now MacOS, which runs on top of Darwin, has a slightly different heritage. It comes from BSD, but I believe it derives from a pre-386BSD branch (although they have incorporated many imporvements from NetBSD, then FreeBSD). It probably it pretty directly decended from LITES, a BSD "personality" which ran ontop of the MACH microkernel. That doesn't necessarily mean that LITES or Darwin is a Microkernel system - rather, they are a "single server" - a single kernel hosted ontop of the MACH subsystems. This means it takes advantage of MACH VM architecture, and probably MACH hardware drivers, but it doesn't take advantage of MACH message passing or other mk technologies.

Much like Windows NT was designed as a microkernel, they ended up defeating the micrkernel architecture to preserve system performance, without wasting previous effort. In this case, the previous effort was from NeXTSTEP, which also ran ontop of MACH. Then Apple incorporated updates to MACH (2.5 to 3.0) IIRC, BSD (4.3 to 4.4) and OpenStep... and every other technology that came since.

There is much more info to google for, but I believe I have thorougly answered your question. Just remember, there is traditional BSD (usually referring to BSD4.4, as opposed to AT&T SYSV) and there is Post-UCB unencumbered BSD (usually 386BSD derived legally unencumbered forks under a "BSD" style Free Software license).

Re:bad news for Linux?

[MobyTurbo]

I have no idea how this got moderated as "insightful". :-)

While it is always good to see new features available on free OSes, I'm concerned that NetBSD's SMP might affect Linux adversely. Linux's advantage over the *BSDs has always been that it has more cutting-edge features like SMP, preemptive kernel threads, strong math emulation, and journaling filesystems. These features are required in enterprise applications, so many sysadmins choose Linux, despite it's cobbled-together nature and lack of good support.

SMP is already in FreeBSD, soon to be improved further. NetBSD is developing theres, and like all NetBSD features is likely to be implemented mature the first time. (Rather than "if it works it's good" NetBSD's philosophy is "if it's good it works") The "strong math emulation" of Linux is already avalilable in FreeBSD, and maybe the other BSDs, I haven't checked their kernel option files lately (another advantage of *BSD, kernel configuration and recompiling is easier), allows you to optionally taint the kernel with GPL code to use the GNU math emulation Linux uses rather than the weaker BSD version. (Linux, of course, is already tainted with GPL code; of course, rather than a bug, this is considered a feature. ;-) ) I dunno why you consider this to be a "cutting edge feature" though, if I recall correctly, this is for 386 and 486sx processors only. :-)

As for journaling filesystems, NetBSD, alone among the BSDs, already offers one; though I'm not sure how mature it is. (The 1.6 release notes say that it's more stable now.) However, soft updates offer many of the advantages of the additional stability journaling filesystems bring. The main reason why journaling filesystems are sutch a big deal on Linux is because ext2 is such an unstable file system, especially with asyncronous metadata, compared to the Berkeley Unix FFS BSD uses. If you don't have asyncronous metadata, yet have the rest asynronous, the whole advantage of redundant metadata storage disappears as I understand it. Of course, you don't have to fsck most of the time with a journaled filesystem; this is the only advantage remaining. In my mind though it is, in practice rather than as a brag, for important 24/7 servers to recover more fast after a power outage; but if you need that kind of availability you should have a power genererator or at least a UPS to deal with the power outage in the first place.

Sadly, we may see a flash flood of business customers moving to NetBSD. I would recommend selling any Debian stock you stil have lying around

I assume you're joking, since Debian is non-profit. NetBSD is owned by the NetBSD foundation, it's also non-profit, though Wasabi and others do manage to run businesses based on it. (After all, it has the Abbie Hoffman of software licenses - the BSD license, "Steal This Code". :-) )

Real SMP? Or does it just "see" more cpu's?

[CoolVibe]

See subject line. AFAICR, x86 "SMP" under NetBSD was only spinup, it could 'see' more cpu's, and spin them up, but it never did anything with the extra cpus. Do they finally have stuff like threading and running processes on more cpus in order? I don't own SMP boxes, but there's a couple of places where I worked that would drool over a stable SMP NetBSD box if that was possible :)

Hmm... Time to scour off the netbsd-current@ mailinglists again for answers...

Re:bad news for Linux?

[Foresto]

Um, even if your premises were solid, I'd still disagree with your conclusion. I think feature competition between Linux and BSD will more likely benefit both operating systems than harm either one.

Re:bad news for Linux?

[dohcvtec]

I have to assume you're speaking facetiously in the voice of a Linux advocate, right? Because in the spirit of free software, people use what works best, and if that happens to be NetBSD instead of Linux, then so be it; there is nothing to be sad or worry over.

a cpu does not a system make

[honold]

you still have io, memory, and chipset performance to improve or bog down your results

that aside, i'd love to see a single metric (really, just one) where a 1.2ghz p3 would get outperformed by 2 underclocked tualatin p3s (to make the competition fair - they'd blow the 600ebs away) on the exact same rig.

2x600mhz != 1.2ghz. it's more around 900mhz average, if you're lucky.