SMP Now In OpenBSD HEAD

A number of people dropped e-mails this morning saying that OpenBSD has now got SMP, according to a post from Niklas Hallqvist.

Comments

No SMP? Huh?

[Anonymous Coward]

Uhh, am I the only one here who was completely unaware that OpenBSD didn't have SMP support?

I guess I've been stuck in the Linux world too long. I never would have guessed this. I haven't been running a single-processor system, workstation or server, since 1998.

The bigger news here, for me, is that Linux just jumped way up on my totem pole of respect.

Hooray!

[jb.hl.com]

Now it can die with super SMP efficiency and at double the speed!

Just kidding, mods don't hurt me :)

Re:Hooray!

[turgid]

Now it can die with super SMP efficiency and at double the speed!

Depending on the implementation, it could only be 1.5 times the original speed. In certain special cases, a good implementation might be able to make it die at 2.1 times the speed or greater.

Re:Hooray!

[Anonymous Coward]

Ahh, heck, 2.1X is nothing--a really bad implementation could just die instantly. :-)

Re:Hooray!

[Anonymous Coward]

Leave poor Windows 9x out of this. We were talking about BSD!

Re:Hooray!

[Anonymous Coward]

Meep meep! MIPS!

Re:No SMP? Huh?

[jcr]

Uhh, am I the only one here who was completely unaware that OpenBSD didn't have SMP support?

Nope, it caught me by suprise, too. Do any of the other BSDs lack SMP support in this day and age?

-jcr

Re:No SMP? Huh?

[Anonymous Coward]

I didn't know about the previous lack of SMP support either. Does anyone know if BSD is capable of supporting IPv6?

userland and config scripts

[tronicum]

I wonder if these CVS changes are enough to support good SMP. At least some userland utilities should support SMP and some of the config scripts that generate make files will probably change (if not autogenerated like automake).

It would be interesting if it scales well so that it really makes sense to use it. The trend goes to some "double core" CPUs, too...

Re:userland and config scripts

[the morgawr]

Could you explain what you mean?

OpenBSD has both a kernel and a userland and this series of merges from the SMP branch (the final one happening this morning) included everything needed to have working SMP. To my knowledge the SMP changes didn't change any user-visible APIs or data structures, so I'm not even seeing why anything in userland would need to be changed.

Uh oh, here comes SCO!

[Anonymous Coward]

BSD could not have become so popular so quickly without IBM taking the SCO code, putting it in Linux where the BSD people took GPL code put it under BSD license and then.... and anyway, this is based on Minix, because no one person could have written BSD, and AT&T probably would sue them for this anyway. This is hurting american economic... GPL is communistic... and Alex de Tocqueville said that... BSD is dying anyway..

Oh man. I just give up.

Re:No SMP? Huh?

[thanasakis]

Yes: ipv6 ports for freebsd

Re:No SMP? Huh?

[Anonymous Coward]

haha nice try, 4.x DOES have smp.

Re:No SMP? Huh?

[BeeRockxs]

OpenBSD has supported IPv6 since version 2.7.

Re:No SMP? Huh?

[jellomizer]

Well OpenBSD has a different set of priorities then Linux. OpenBSD philosophy is to do it and do it well make absolutely sure that it is secure. This strong focus on security slows down a lot of development and thus keeping OpenBSD from the leading edge of technology. Now that a lot of SMP technology has matured and proven its worth it is now time for an OpenBSD implementation. Being on the leading edge is nice. But when you have a solution that must absolutely has to be running and secure there is no shame on being a little behind the times in technology.

Re:No SMP? Huh?

[life4m]

No, SMP was only absent in OpenBSD. It was just not a priority for the project. Nevertheless, it is a welcome addition.

Can anyone "in the know" shed some light about the qualities of the OpenBSD code? How does OpenBSD's SMP model compare to fine-grained locking, such as in FreeBSD-CURRENT?

When I see how much effort and trouble has gone and is going into locking down the FreeBSD kernel, I am guessing that OpenBSD's SMP support will be fairly primitive to start with. Or are there heavy ports from FreeBSD?

Re:No SMP? Huh?

[Morth]

*BSD had (useful) IPv6 long before Linux thanks to kame. OpenBSD is also the last of them to get SMP support, even if it's pretty fresh in NetBSD too (a year or so).

Re:No SMP? Huh?

[Anonymous Coward]

I was under the impression that SMP support has been in FreeBSD for a long time, since the 3.x days at least. It wasn't brilliant SMP (lots of coarse grained locks for example) but it was SMP. Fine grained SMP and new SMP code has emerged with FreeBSD 5.x however.

Re:No SMP? Huh?

[Anonymous Coward]

Sure, *BSD was the first OSes to implement IPv6, and FreeBSD have IPv6 enabled by default.
And for SMP, FreeBSD and NetBSD have had this feature for a long time.
Personnaly i remember removing this line in my freebsd kernel config file for almost 3 years.

Re:No SMP? Huh?

[Shakrai]

haha nice try, 4.x DOES have smp.

I have mod points so I actually researched the AC to see if he was right or wrong. Rather then waste them on an AC though I figured I'd actually post a link to back up his claim: FreeBSD Symmetric MultiProcessor Kernel.

Re:No SMP? Huh?

[mikael_j]

Nice troll, FreeBSD 4.x does have SMP although it's pretty inefficient. 5.x has much better SMP but it's currently labeled "new technology release" and they recommend 4.x for production servers. 5.x is quite stable though..

/Mikael

Re:Time to move openbsd.org to OpenBSD then ...

[kunudo]

*cough* hotmail *cough*

Re:No SMP? Huh?

[jmartinp]

FreeBSD 4.x has SMP and if memory serves me, 3.x had SMP as well. It's true that the 5.x branch includes a reworked SMP, also known as SMPng, which includes more fine-grained locking, but saying that 4.x branch lacks SMP is misleading to say at least.

Re:No SMP? Huh?

[Anonymous Coward]

Can I have some of what you're smoking?

FreeBSD 4.9-STABLE #7: Fri Mar 12 15:22:05 EST 2004
xxxxxxx@xxx.xxxxxxxx.xxx:/usr/src/sys/compile/DERW OODSMP
Timecounter "i8254" frequency 1193182 Hz
CPU: Pentium III/Pentium III Xeon/Celeron (497.44-MHz 686-class CPU)
Origin = "GenuineIntel" Id = 0x673 Stepping = 3
Features=0x383fbff<FPU,VME,DE,PSE,TSC,MSR,PAE,MCE, CX8,APIC,SEP,MTRR,PGE,MCA,CMOV,PAT,PSE36,MMX,FXSR, SSE>
real memory = 1073733632 (1048568K bytes)
avail memory = 1039605760 (1015240K bytes)
Changing APIC ID for IO APIC #0 from 0 to 2 on chip
Programming 24 pins in IOAPIC #0
IOAPIC #0 intpin 2 -> irq 0
FreeBSD/SMP: Multiprocessor motherboard: 2 CPUs
cpu0 (BSP): apic id: 0, version: 0x00040011, at 0xfee00000
cpu1 (AP): apic id: 1, version: 0x00040011, at 0xfee00000
io0 (APIC): apic id: 2, version: 0x00170011, at 0xfec00000

co-incidence? I think not.

[Anonymous Coward]

The fortune I got at the bottom of this page is:

Dying is one of the few things that can be done as easily lying down. -- Woody Allen

Re:Time to move openbsd.org to OpenBSD then ...

[Anonymous Coward]

That's actually crap. They use the university of alberta's server as it has more bandwidth available to it than their own connection.

Re:No SMP? Huh?

[Octorian]

Also, I belive that NetBSD already has SMP support (not sure when it went in there, though). So does Darwin, of course.

Re:Time to move openbsd.org to OpenBSD then ...

[zyche]

RTFFAQ.
http://openbsd.org/faq/faq8.html#wwwsolar is

"www.openbsd.org and the main OpenBSD ftp site are hosted at a SunSITE at the University of Alberta, Canada. These sites are hosted on a large Sun system, which has access to lots of storage space and Internet bandwidth. The presence of the SunSITE gives the OpenBSD group access to this bandwidth. This is why the main site runs here."

And even given this, ftp.openbsd.org is usually very slow around release time.

Re:No SMP? Huh?

[Korpo]

I welcome your focus on security, but "being a little behind the times in technology"???

OpenBSD may be the safest web server around. But without SMP it doesn't suit the servers needing the most protection, does it? Nearly every "serious" web server is SMP - at least in corporations, or am I wrong?

Having said that, it would be interesting how long it will be till SMP is considered a safe and stable feature in OpenBSD (else it would not be worthwhile, according to your comment - and I guess I agree there)?

The OpenBSD project simply has too few resources, and that's for sure! And that has nothing to do with "different set of priorities then Linux". I'm pretty sure, that Linux developers are numerous enough to aim for all three goals: reliability, safety and efficiency.

Multiple Niches

[Mark_MF-WN]

Yeah, the whole BSD vs Linux conflict is ridiculous. There are a lot of niches out there in the software world, and Linux can't fill them ALL. Linux is nice on the desktop, handheld, and cluster, but the BSDs seem well suited for firewalls, routers, and other kinds of always-on equipment. OpenBSD in particular seems useful for bastion hosts, because of its rock-solid security. And of course, we still need Windows for hardcore gaming.

The point? Niches -- there are a bunch of them. Although I'm a loyal Linux user, I love the OpenBSD project. It contributes a great deal of useful software and bugfixes that help the whole community.

Re:Multiple Niches

[drinkypoo]

I beg to differ. Linux can fill them all - at least, it can go anywhere OpenBSD, netbsd, or FreeBSD can go. Whether it's a better tool for the job or not is a subject I will leave to a future flame war (and of course, many past ones.) Linux has been shown to scale down to systems with no FPU (a limited version admittedly) and up to systems with many processors thanks to the work of IBM and SGI.

OpenBSD has better code review and often employs better coding practice than linux, I will grant you, but with the addition of grsecurity and the proper use of firewalling, Linux is no security slouch either.

And, windows is not a better platform for gaming than linux in any way other than availability of games. Many games run faster on linux than on Windows. In general it's an issue of what platform it's optimized for.

Linux has many features desirable for routing packets that are missing or not as complete in other operating systems.

Re:No SMP? Huh?

[arivanov]

4.x has SMP. Used it for a while before switching to 5.x.

4,X SMP is not the best performer out there, because the locking is not very fine grained, but it works.

Re:No SMP? Huh?

[Anonymous Coward]

Does anyone use ipv6?

Re:No SMP? Huh?

[AKnightCowboy]

*BSD had (useful) IPv6 long before Linux thanks to kame. OpenBSD is also the last of them to get SMP support, even if it's pretty fresh in NetBSD too (a year or so).

That's like saying your car brand was the first to support running on both gasoline and optionally hydrogen. Good for you, but let me know when there's a hydrogen fueling station in my city, nay, my entire state, and I'll happily switch to your brand. IPv6 is irrelevent right now outside of lab environments and isolated networks (cell phone providers, etc.).

SMP defined

[Anonymous Coward]

Wikipedia article on Symmetric multiprocessing (SMP).

Re:SMP defined

[Anonymous Coward]

Thanks, I didn't have a clue what SMP stood for. They really need to stop assuming everyone who reads /. knows it all.

Re:No SMP? Huh?

[dekeji]

Now that a lot of SMP technology has matured and proven its worth

You've got to be kidding: "SMP technology" proved its worth decades ago.

OpenBSD philosophy is to do it and do it well make absolutely sure that it is secure.

By itself, a concern over security does not mean that you have long time-to-market for a product. It is possible to implement secure software systems rapidly, but you have to give up on other things (like, for example, an insistence on manual storage management and raw pointer manipulation).

The degree to which OpenBSD's tradeoffs are desired by the market becomes clear from how widely it is used--and I don't mean that cynically: OpenBSD fulfills a significant function for a small market, but most people choose something else. It's good that we have choices like OpenBSD, even if most people choose not to use it.

Re:No SMP? Huh?

[Tranzig]

The bigger news here, for me, is that Linux just jumped way up on my totem pole of respect.

I wonder where Windows NT is on your totem pole, because it had SMP support years before Linux 2.0. And ACPI support and journalizing filesystem support and modules (drivers) support and so on...
I know I will be modded down for such blasphemy.

Anyways OpenBSD has (at least had last year) scalability issues, it scales pretty bad, and it needs to be solved ot get SMP really effective.

yayy

[xyloplax]

Now I can weld a new proc onto my sparc 5 firewall!

Re:yayy

[0racle]

Good luck, Just built it on a dual hypersparc ss10, only sees CPU0

Wave of the future

[ObsessiveMathsFreak]

Ultimatly all PCs will have to move to multi-processor systems ala G4,G5, so OpenBSD was just making a locical progression that Linux made back in the 2.0 kernel was it? I'm hoping more multiprocessor machines will come out in the future.

Dispel my ignorence please.

Does windows have SMP/NUMA support yet?

Is it economically more feasable to build a multi- (not so great)processor machine rather than a powerful single CPU?

Thanking (kissing upto) you in advance for the answers.

Re:Wave of the future

[REBloomfield]

Does windows have SMP/NUMA support yet?

Yes.

Is it economically more feasable to build a multi- (not so great)processor machine rather than a powerful single CPU?

No.

Re:Wave of the future

[olderchurch]

Please mod parent up with +2 simplicity ;)

Re:Wave of the future

[mattyrobinson69]

windows does have SMP support, but you have to pay through the nose for it. i think its either a server license or an smp server license, im too lazy to google for it

Re:Wave of the future

[TheRaven64]

Does windows have SMP/NUMA support yet?

Windows NT has always had SMP support (I think, I haven't used the 3.x series). The workstation version supports 1-2 CPUs, the server ones support more, although they are limited to 16 (I think. Maybe 32. Too lazy to check) due to a the size of the processor table (see Tanenbaum's Modern Operating Systems for more information) Is it economically more feasable to build a multi- (not so great)processor machine rather than a powerful single CPU?

More or less. Top of the line CPUs usually carry a large premium, and the mid-range ones usually have a much better price/performance ratio. On the other hand SMP motherboards are usually more expensive. In the future it will become more sensible, since the cost of increasing the speed of a CPU by n% increases over time.

Re:Wave of the future

[cpeterso]

yes, Windows NT has had SMP support since 1988, when NT development began. NT also had an O(1) process scheduler since 1988, but Linux likes to accept credit for this "invention" 15 years later..

Re:Wave of the future

[Anonymous Coward]

I don't understand what's so hard about this (O(1) scheduler). It's called a fucking linked list! You maintain a head, and a tail. Tasks go in on one end, come out on the other end. Adding? NewTask.Next = Head; Head=NewTask; Removing? Result=Tail; Tail=Tail.Next; (w/ a sentinel value instead of NULL for the end, where the sintenel points to it's self, and THAT'S IT! O(1)!).

Want multiple priorities? Have multiple linked lists. Still scales linearly w/ the number of tasks. What I want to know is WTF did Linux have before it had a O(1) scheduler? How did it suck that badly? And how did it manage to do it for what, 1.0, 1.2, 2.0, and 2.4? 4 versions?

And finally what's great is how the O(1) scheduler is so great, but Linux was so great before too! Windows sucked compared to Linux before when Linux was interior, and now Windows sucks more! Ahh, the joys of those who know just enough...

And finally the one thing that never comes up w.r.t. O(1) scheduler is it doesn't apply to sleeps. I'd like to see someone implement a O(1) sleep algorithm for multiple processes sleeping. You've gotta have a priority queue somewhere... And that essentially means for any non-infinite wait you're not O(1). Period...

Re:Wave of the future

[Anonymous Coward]

True, but some of us have old systems laying around now.

I just was "given" a dual CPU box for a project I was doing. I wanted to put OpenBSD on it but I did not want the "other" cpu to go to waste. So hello Linux.

Re:Wave of the future

[drsmithy]

Does windows have SMP/NUMA support yet?

For over a decade now.

Is it economically more feasable to build a multi- (not so great)processor machine rather than a powerful single CPU?

Depends on your objectives and restraints.

Re:Wave of the future

[trashme]

Does windows have SMP/NUMA support yet?

Do not lump these two terms together. Supporting NUMA can be much more complicated than SMP. Specifically, the difficulty is maintaining cache coherency. Check out the Wiki link for more info.

Re:What's SMP?

[Anonymous Coward]

That's what you get for being an OpenBSD user!

Hehe. Sorry.

Faulty Assumption?

[SteveM]

They really need to stop assuming everyone who reads /. knows it all.

They don't assume you know it all. They do assume you are smart enough to do some research and find info on stuff you don't understand.

How hard is it to google "SMP"?

SteveM

Re:Faulty Assumption?

[subStance]

Very difficult, but only when you're trolling for mod points, by writing a bot that looks up a keyword and posts a Wikipedia link, followed by a comment that says thanks for posting the link.

Ignore the bot.

Re:Faulty Assumption?

[DAldredge]

You must not read the Ask /. section of /.

Re:No SMP? Huh?

[Anonymous Coward]

Yes, it's true that NT was there early. OS/2 SMP was due for release in 1993 (before Microsoft and IBM split ways and Microsoft turned it into Windows NT). But there is a bit of a difference in that Microsoft had quite a bit of hand-holding from Intel to develop an SMP solution.

For Linux, by contrast, SMP support was done the hard way through reverse-engeering, experimentation, and careful studying of available technical documents with the end result a stable, scalable SMP solution. To me, that is more impressive.

OpenBSD commands respect...

[emil]

...quite easily. Examine Red Hat's errata list for AS3, then look at OpenBSD's errata. I assume that you will see a rather conspicuous difference in the quantity of changes?

Granted, this list is not entirely fair, as many ports and packages have bug fixes, which would push up OpenBSD's count. However, OpenBSD includes a great deal in the base distribution (SSH, Apache, Sendmail, etc.) that comprises what they assert to be audited, secure code.

To me, the ability to deploy a server and then spend minimal effort with security patches is more important than SMP. YMMV.

Re:OpenBSD commands respect...

[CoolGuySteve]

True, except the patches that do exist for OpenBSD are a pain in the ass to install. They're source only last time I checked and while some people linked from google did have pre-compiled ones, I couldn't find anything that was mature or that even had a working server. If there's somewhere where sparc64 tar files exist, I'd really like to know.

I understand their resources are limited, but come on, the University of Alberta can't even afford ONE $200 machine on which to compile these things?

Re:OpenBSD commands respect...

[lysander]

They're source-only updates on purpose -- the only binaries that are available are through releases and (recently) snapshots.

Once you update your src tree, you can recompile the kernel and system binaries if you need to -- whatever the patch requires. As a bonus, since everything is backed by cvs, you could make local changes and merge in updates rather seamlessly.

The snapshot builds, I think, are taken from -current rather than -stable.

Re:OpenBSD commands respect...

[Anonymous Coward]

But maybe I prefer not to keep a compiler on my firewall. That does make deploying new patches a bit of a pain since they have to be compiled on a separate machine. Fortunately, it's rarely needed.

Re:OpenBSD commands respect...

[CoolGuySteve]

Exactly, there are too many variables to have it be done reliably by a cron job on another machine unless you dedicate that system to running OpenBSD. When something is no longer automatic, it tends to get put off for long periods of time in smaller or understaffed departments.

The cvs excuse seems pretty lame, I bet most people do run the base setup unless they want linux compatibility. Even then, as I said, it's not hard to buy a $200 machine to compile a handful of the most popular config permutations. Just a tarball containing all the binary patches up to that point would be leaps and bounds over what exists. And really, the reason people rely on OpenBSD in the first place for their servers is so they can forget about it, this is the only fly in the ointment.

Re:OpenBSD commands respect...

[macdaddy]

That's like comparing a Datsun to a GMC. The Datsun is utterly barren and featureless (or was when it was introduced). Just because O-BSD has few security patches doesn't mean it's better. It has no features either.

Re:OpenBSD commands respect...

[Anonymous Coward]

Yeah because a webserver and a mailserver don't do much for you.

Re:OpenBSD commands respect...

[Anonymous Coward]

Then buy one,set it up, and announce it. Don't forget to include archs other than i386. So you'll need: i386, amd64, sparc, sparc64, cats, alpha, vax, macppc, and m68k. At least (I'm probably forgetting an arch or two). Put them on the net, compile patches, and make them available for everyone.

SMP is ...

[Anonymous Coward]

SMP is Soviet Magnetic Pulse. Much better than the imperialist running dog american's electro magnetic pulse.

And we proved it by running simulation code aquired by KGB and running on our dual 4.7 MHz 8088 CPU symmetric multi processing computer.

Great news

[karmawarrior]

I know I'm not the only one who's felt that SMP may make or break whether you can efficiently deploy OpenBSD on your workstations and servers. SMP may not be high priority for security, but a secure OS that only gives you access to half the processing power of your system is always going to be a disappointment. And SMP, while not a priority, does help security in that it makes DoS attacks much more difficult.

(Why? Because many typical DoS attacks work on the basis of a single process hogging the CPU. If you have two, a system administrator can log in and kill the process. If you have one, that system administrator will find it more difficult to do so. It's not a cure-all, but it helps.)

In all, excellent news. Thanks to everyone who made this possible, regardless of whether you just coded, or you campaigned for support for the OpenBSD project.

Re:Great news

[Anonymous Coward]

In what way was it rubbish? It looks factually correct to me.

Re:Great news

[cmowire]

Actually, the OpenBSD folks held off on SMP support for a while for a good reason.

Why? Simple. It's a new avenue for bugs and exploits because it makes race conditions more likely to occur. Remember, OpenBSD has a reputation to uphold. ;)

Re:Great news

[evilviper]

many typical DoS attacks work on the basis of a single process hogging the CPU. If you have two, a system administrator can log in and kill the process. If you have one, that system administrator will find it more difficult to do so. It's not a cure-all, but it helps.

Well then, why not have a 128 processor system, so when 127 processes are tying up a processor a piece, you can still log-in and kill them...

SMP should not be used as an alternative to failover, and not as a half-assed fix for DoS issues.

Got a CPU hog? Have a watchdog program running (at nice -19) to kill/restart any programs that are using up much more than their fair share of resources.

Re:No SMP? Huh?

[Morth]

Woah, a little less zeal thanks. Apparently the question mattered to my parent, since (s)he asked about it...

Re:No SMP? Huh?

[Chreo]

Perhaps you should do your research a bit more thoroughly next time. The link you are referring to is about FreeBSD SMPng (ng = next generation, implying a previous generation). SMPng is a core tech of FreeBSD 5.x but SMP has been a feature for 4.x a long time. From the page you linked:

FreeBSD supports Symmetric MultiProcessor kernels in the following releases:

* 3.0-RELEASE/STABLE
* 4.0-RELEASE/STABLE
* 5.0-CURRENT

Get informed plix...

[CptnHarlock]

OpenBSD is NOT premirely a webserever. It can and is used as one (I do it), but its main use is as a firewall. It even got voted as "best application for firewalling" by the MS crowd.. :)

I've been aware of OpenBSDs lack of SMP since 2000 when I installed it on a double PentiumPro 180 Mhz. I Thought it was weird it didn't have SMP since Linux and FreeBSD had it but after some reading I accepted the desision.

I guess the OpenBSD guys went for SMP considering the double core desktop processors gaining more ground. Am I correct?

Cheers...

Re:Get informed plix...

[drinkypoo]

The OpenBSD guys went for SMP because more people were bitching about it for whatever reason - but that is only secondary - primarily because one of them must have either wanted it for personal use, or just felt like working on it. OpenBSD is primarily Theo-driven, and after that, it's primarily other-core-developers-driven.

The implementation of SMP is likely not based on the dual-core processor thing but just the fact that SMP systems are so cheap now. You can pick up dual P2 systems used for a song now, and they make nice routers if your OS supports them, because while your kernel monopolizes one processor, other things can still be executing on the other. A complete network appliance is after all not just a kernel, there are userspace programs which are handling other ancillary tasks beyond just routing packets.

Re:Get informed plix...

[jimi1283]

Ok, I have mod points and I have to comment on this. I can say without a shadow of a doubt that the OpenBSD team does not do anything at the whim of the bitching masses. Did you not read the recent mailing list thread about the decision to stop updating Apache in the ports tree??? And what about when they removed ipf???

Theo may be a hard headed son of a bitch but he takes care of buisness and is not swayed by popular opinion. This is not a sudden development, there were articles a year ago about how the OpenBSD team was looking for a full time programmer to work on SMP.

This has been planned, and when OpenBSD executes a plan they execute it well.

Re:Get informed plix...

[drinkypoo]

Everyone is swayed by the bitching masses. Sometimes, they're swayed in the direction away from what they want. Starting when I was a child I have been this way. When my mom would harp on my to do my chores on her schedule I'd end up not doing them. I am a contrary bastard and always have been :)

Thank you for commenting instead of modding. I far prefer comments to moderation.

Re:Get informed plix...

[Anonymous Coward]

This work was done because a company contracted a developer to do it.

Not so fast...

[tqbf]

The initiative to ADD SMP to OpenBSD (and the announcement that "a full time developer was working on it") occurred less than 4 months ago. It took FreeBSD years to get SMP "right", during an adolesence where stability seriously suffered.

Neither NetBSD nor OpenBSD have a significant fraction of the user base of FreeBSD (and an infinitessimal share compared to Linux), and neither NetBSD nor OpenBSD have comparably-sized development teams. FreeBSD SMP had antecedants to build on as well. So logic dictates that it should take longer for OpenBSD SMP to be "ready".

OpenBSD went through a comparable architecture change when they swapped virtual memory systems a few years ago, and several subsequent major releases of OpenBSD had serious VM stability problems (many of them synchronization issues). SMP is even harder (and more of it involves synchronization).

So, is there some mitigating factor here that would convince anyone who was paying attention to deploy a mission-critical system on SMP OpenBSD in 2004?

Re:Not so fast...

[cubidou]

Since OpenBSD SMP support is mostly NetBSD's plus code style changes, I'd say it only took 4 months to import what was started 3 or 4 years ago by Bill Studenmund for NetBSD.

Credit where it's due.

Re:Not so fast...

[cubidou]

Since OpenBSD SMP support is mostly NetBSD's plus code style changes, I'd say it only took 4 months to import what was started 3 or 4 years ago by Bill Studenmund for NetBSD.

Oh, and I forgot, removal of all ACPI code, too.

That means OpenBSD won't work with MP computers that have a broken MPBIOS or simply require MPACPI (like most, if not all, HyperThreading processors do).

Re:Not so fast...

[Nimrangul]

So, is there some mitigating factor here that would convince anyone who was paying attention to deploy a mission-critical system on SMP OpenBSD in 2004? If you see it in the 3.6 release, consider it ready. If you don't then it wasn't ready yet.

Re:Not so fast...

[LizardKing]

NetBSD nor OpenBSD have comparably-sized development teams [compared to Linux]

That's not as accurate as you might think. If you look at Linux kernel development, there is a core of really talented developers who work on the guts - the VM system, VFS, threading, etc. Most other people only poke away at the edges, driver writing for instance. If you compare the number of core developers working on Linux and the NetBSD kernel, a similar number of names crop up with regularity on the relevant mailing lists. NetBSD also has a strong following in academia, and gets a lot of input from talented people with plenty of time to tinker. This partly compensates for the lower number of people working professionally on NetBSD (mostly Wasabi employees) compared to Linux. The BSD's also started from an almost complete Unix implementation (4.4BSD Lite), while Linux had to reinvent the wheel a few times.

Chris

Re:Not so fast...

[kivaapina]

It was Bill Sommerfeld. Get the facts straight ;)

Re:Not so fast...

[cubidou]

We have too many Bill S. :)

Re:Not so fast...

[the morgawr]

That means OpenBSD won't work with MP computers that have a broken MPBIOS or simply require MPACPI (like most, if not all, HyperThreading processors do).

If you'd read more of the archives linked from the article you would have seens that they are working on it at the Hack-a-thon.

Also as I'm sure you know, most of the work in implementing an SMP system is not in writing lock.c but in going through the entire OS code (which for Open looks only kinda-sorta like Net at this point) and using the lock and debugging race conditions. I think the OpenBSD SMP CVS logs will bear this out.

Re:Not so fast...

[the morgawr]

A good chuck on the basic SMP code was done in the SMP branch a while back. Saying it took only four months is misleading.

Re:Not so fast...

[Anonymous Coward]

No you're flat wrong.

*Most* of the work in a single-threaded kernel involves the architecture code and the scheduler.

Locking entry points into the kernel is pretty trivial by comparison.

Re:Not so fast...

[Anonymous Coward]

Are you joking? You are either blatantly lying or unexcusably ignorant.

I think you've been in jail too long.

Re:Not so fast...

[evilviper]

OpenBSD went through a comparable architecture change when they swapped virtual memory systems a few years ago, and several subsequent major releases of OpenBSD had serious VM stability problems

I was using OpenBSD at the time, and never had the problems myself. I wasn't following the mailing lists, so I didn't even know there were serious problems.

So, is there some mitigating factor here that would convince anyone who was paying attention to deploy a mission-critical system on SMP OpenBSD in 2004?

Yes. Their track record. If it's not stable about a month before the release, expect it to be disabled and/or removed. If it is reasonably stable, it's going to go through a couple weeks of nothing but bug testing and fixing before it's released. OpenBSD releases are nothing like Linux releases, you aren't going to have to download the new version a day later because there's a show-stopping bug in the version they just released...

That said, even though they are putting in SMP support, I wouldn't recomend switching soon. Even if it is stable and secure, don't expect performance to be mind-boggling for a while.

Re:What's SMP?

[tehcyder]

symmetric multiprocessing...I believe M$ invented it for Windows NT ;-)

Re:No SMP? Huh?

[Roofus]

Don't bother wasting your time replying to this guy. I mean, look at his name for Christ's sake - MS_is_the_best

Re:No SMP? Huh?

[Sancho]

AC says: FreeBSD 4.0 DOES have SMP.
Reply says: I wanted to post this link to back up his claim.
You: I will refute you by proving your point.

Everyone from the AC down the thread was saying that FreeBSD 4.0 had SMP.;)

Re:No SMP? Huh?

[Alioth]

OpenBSD's focus is vastly different from Linux's (I use both, and I use them where appropriate). OpenBSD's primary focus has been on security and correctness of the implementation. Compare OpenBSD's pf with Linux's iptables - pf is so much more powerful and useful than iptables the difference is like night and day. The OpenBSD pf has security features that cost large sums of money in the closed source world.

SMP simply is not a priority for OpenBSD. The kind of uses OpenBSD is put to hardly ever requires it, so it's not in the least surprising that they are only just implementing it now.

Can't BSD people be polite in subject lines?

[Korpo]

So, to be more general, we have more uses, but except firewalling, all of them reside on the server side.

All are services, virtually none workstation-related. While I'm aware, that OpenBSD makes a workstation, too, of course, I'm guessing it really excels as being a secure service provider, be it server or firewall/gateway.

Having said that, there's still the problem of production use in larger environments (universities e.g.), and I'd guess there has been need for SMP in terms computing power for quite some time.

How did they solve that till now (and for the forseeable future, till SMP gets stable)? Clustering?

Re:Can't BSD people be polite in subject lines?

[CptnHarlock]

Was I impolite? Sorry, I even re-edited my first subjet that includede "OMFG" & "FFS".. j/k.. ;)

For me OpenBSD has been a OS where I've been willing to give up performance for security. This is no major problem considering that todays (and even yesterdays) computers are extremely efficient if you don't shove a lot of fancy graphics down their throat. My fw at home was 486sx25 and got substituted by a P100 only because the P100 was far more silent. (oh, and it runs FreeSCO)

Speaking of performance, I did get very very happy with 2.9 wich boasted a 60x filesystem I/O boost.. :) ..

Anyway, I have no experience of huge enviroments. I've worked at smaller schools with a maximum of around 250 desktops/servers. I used to work at a school around 2000/2001 with some 150 desktops and all of them with real IPs (no fw). Most of them had servers running since it was a webdeveloper/designer school. Some computers were Macs but most run NT4. Needless to say they got haxxored more and more. I got fed up with it and started running RedHat 6.2 at my desktop (as I did at home) but even here I got fed up with checking for patches and checking logs for intrusions so I went for OpenBSD at the desktop and we started moving all exept the .asp pages to a second OpenBSD 2.8 computer. Never had any problems with either of these comps exept an ftp patch and an ssh patch IIRC. Our fileserver was a Samba wielding Slackware Linux-box and we also set up a fw, don't remember if it was Linux or OpenBSD though.

As you see I am an not "BSD people" but rather an OS agnostic (aka OS whore) and believe that all OSes have a good place in the matrix. OpenBSD is my choice for "deploy and (almost) forget". Security issuses come so seldom that they get posted on /. which I check daily.. ;)

BTW.. The other day I celebrated the 777 days uptime (that's 2+ years) of a OpenBSD 3.0 machine.. :')

Cheers...

Re:No SMP? Huh?

[Dwonis]

Yep. It's great for working around the limitations of NAT, among other things. Especially with 6to4.

soekris

[curator_thew]

Read the soekris website, it puts it nicely:

FreeBSD The most powerful x86 open source Unix OpenBSD The most secure open source Unix available NetBSD The most portable open source Unix available Linux The most popular open source Unix

Re:soekris

[Anonymous Coward]

Numbers please.

Re:No SMP? Huh?

[curator_thew]

You are an idiot: IPv6 support is a mandatory requirement to have as IPv6 rollout comes ahead in the next few years. It's a testimony to the solid reliability of the BSD protocol suite that the main experimental development for IPv6 (notably via. KAME) was undertaken on BSD. So this forward thinking will pay off.

it's because we're dying

[Anonymous Coward]

Can't BSD people be polite in subject lines?

No we can't. You know how people with terminal deseases get... ;)

Re:No SMP? Huh?

[Chreo]

Actually, it was a mistake caused by being tired. I misread the parent and thought he was supporting the claim that only 5.x had SMP, since the linked page was on the topic of SMPng. The only good deed I did (with my mistake and all) was posting the quote from the page showing that even 3.x had SMP since not to many bother viewing links.

W2k Pro supports 2 procs

[FatSean]

I'm sure XP costs more. I'm sticking with 2k.

Re:No SMP? Huh?

[LizardKing]

Do any of the other BSDs lack SMP support in this day and age?

NetBSD has had SMP support for some time, and it is now working on a number of platforms (sparc, i386, etc). It's largely been the work of Bill Studenmund, and also formed the basis of OpenBSD's SMP support. This kind of sharing is quite common amongst the BSD's, especially Open and Net.

Chris

SMP didn't make it more secure, so not a priority

[Anonymous Coward]

I find that most changes to openbsd answer the question "how can we make this more secure?"
SMP is a scalability issue, so it did not receive the developer time needed to make it work over the many many platforms that OpenBSD is ported to.

Imagine if the OpenBSD people had spent time on SMP instead of OpenSSH. We all benefit from the focus on security even if we don't use OpenBSD.

Re:Time to move openbsd.org to OpenBSD then ...

[Tony-A]

Looks a bit bad with an OS-maker not using it's own OS for it's website.

Rubbish.
Assuming security is actually a priority, there are some good reasons to run the web and ftp servers on something different. One or the other might have problems occasionally, but the odds of both having problems at the same time is (considerably) less than the odds of either one of them having problems.

(Of course is may just be the storage and bandwidth available from the University of Alberta;)

Re:No SMP? Huh?

[fimbulvetr]

No, you are wrong. Google uses clusters of smaller machines. Dual CPU Webservers are generally considered bad, as they don't scale near as well as 2+ webservers serving files via nfs or other technology.
With 2x cpus, when a server gets bogged, it's gonna quickly hit a ceiling if you keep adding processors. Besides, unless you're doing huge engineering apps in real time, a webserver really only needs fast I/O and more memory to grow.

I tend to think it's because of Theo, not everyone gets along with him, if ya know what I mean. It will take a while before SMP is considered secure, as smp introduces quite a few possibilities for exploits, just due to the nature of sharing procs across processors.

Binary patches for OpenBSD...

[emil]

... discussion for i386 and Sparc64. Not endorsed by OpenBSD.org.

Re:No SMP? Huh?

[TheRaven64]

The pretty inefficient is highly dependent on the tasks being performed. For a while I was running a dual P3 700 (512MB of PC133) and a 1.4GHz Athlon (512MB of DDR266), and the dual P3 felt a lot more responsive (both running FreeBSD 4.x).

The inefficiency in the SMP in FreeBSD 5 is due to the fact that all of the kernel uses a single mutex (referred to as giant), so no two processes can be executing system calls at the same time. If you have a process that is doing something outside of a system call then it can still be running on another CPU. That said, the 5.x series has much finer grained locking and considerably better performance (especially when the ULE scheduler is factored in).

Didn't Theo said this was not going to happen?

[Anonymous Coward]

what changed?

Is SMP a good thing towards security?

Re:No SMP? Huh?

[Anonymous Coward]

By itself, a concern over security does not mean that you have long time-to-market for a product. [...] most people choose not to use it.

Yes, and that's the thing: budget scales with popularity.

The primary reason why OpenBSD has gone without SMP for so long is that they didn't have the resources to do it well. Not enough spare people, not enough hardware to test on.

It is possible to implement secure software systems rapidly, but you have to give up on other things

fast, cheap, good.

Re:No SMP? Huh?

[Anonymous Coward]

If you put OpenBSD infront of the Linux box as it's firewall, you have a reasonably secure setup. OpenBSD's first goal is secure and it is not a "Windows" chaser like Linux.

The best part, is if you know the Linux command line, then you can adapt between the two quite quickly.

But if you are a dyehard BSD fan, FreeBSD has had SMP for much longer.

Re:No SMP? Huh?

[Caligari]

I'm pretty sure, that Linux developers are numerous enough to aim for all three goals: reliability, safety and efficiency.

While I like Linux and use it in some situations, I can tell you for sure that most distributions are far from competing with OpenBSD in terms of safety. You are right in saying that OpenBSD has a lot less resources than Linux, but they use their resources in a far more focused way.

• Yes, there are 3rd party patches which hack many anti-buffer overflow protections into the Linux kernel, similar to what OpenBSD has.

• Yes, there is a stateful firewall for Linux.

• Yes, there is ipv6 support for Linux.

But OpenBSD takes all of these things, which under Linux can be half baked and kludged, and packages them together as a polished, stable end product. Their PF work is quite frankly amazing. The features and documentation are unbeatable. Checkpoint and Cisco, watch out!

The key difference between GNU/Linux and the various BSDs is integration. The BSDs assure you that the various things will play together properly. Features are added more conservatively, but they are going to work. The system as a whole is stable.

You know that for example the buffer overflow protections are not going to break half your userland applications, because it has been thouroughly tested on the system as a whole. Some example results of this:

• You know that the packet filter will play nice with the IPv6 subsystem.
• You know that Systrace will work on an SMP kernel.

You also don't get silly things like stable kernels which corrupt your filesystem or ripping out the virtual memory subsystem in a stable kernel and completely changing it.

All these things are very nice when you are running serious production servers.

Linux can perform a large number of roles adequately.

OpenBSD can perform a smaller number of roles excellently

Re:No SMP? Huh?

[Just Some Guy]

IPv6 is irrelevent right now outside of lab environments and isolated networks (cell phone providers, etc.).

I run it as a first-class transport on my network. All of the major services I provide are available over IPv6, including DNS, SMTP, HTTP, HTTPS, IMAP, SSH, and Kerberos; the notable exception being Jabber because I haven't had the chance to upgrade it yet.

More to the point, I'm actually seeing traffic on those services. They're not just some toy implementation floating out in space, but real websites that are getting hits every day and a mailserver that's seeing an increasing amount of IPv6 traffic.

You may not be using it yet, and that's perfectly fine, but don't take that to mean that noone is.

Re:No SMP? Huh?

[Anonymous Coward]

[It is possible to implement secure software systems rapidly, but you have to give up on other things] fast, cheap, good.

I would make the argument that you don't have to give up on any of those.

Keep in mind that the same argument you make for why OpenBSD is behind Linux applies even more to new (and, I would argue, better) approaches to security. OpenBSD may have fewer resources than Linux, but it started off with a complete working codebase and a dedicated developer community. That gives it a huge advantage in terms of resources over more modern approaches to operating system design and implementation, even if those newer approaches were technically superior to it.

Or, to put it more bluntly, OpenBSD users complaining about Linux taking away resources from them is like the pot calling the kettle black: from the point of modern OS design, both OpenBSD and Linux are holding us back.

Re:No SMP? Huh?

[Anonymous Coward]

As a deviout linux user zealist I can safely safe that windows NT rocks. Its the only version of windows i had any respect for, and its the only version of windows I'd look at deploying for legacy mission critical windows applications. I found it simple, easy to use, powerful and reliable. I cant say the same for there more modern operating systems.

Re:No SMP? Huh?

[Korpo]

Thanks!

This was actually instructive and informative.

OS agnostics seem to be polite ;)

[Korpo]

Thanks, that was funny and informative (and this will be moderated Redundant).

OpenBSD does surely profit from the fact it was security-reviewed that much - and is still. So there is more stable, bug-free code, that doesn't need much changing. And won't changed often, because security comes first.

Linux, on the other hand, is massively performance-reviewed, or at least I get that impression. This induces many small changes that impact performance and features, and I guess this will impact security in the long run. Many small changes are hard to track for security reviewers, or at least I think they are.

My workstations all run some versions of Debian, and out of laziness, my firewall, too. If you've got book-shelf full of Linux books you get less tempted experimenting with OpenBSD, I guess. ;)

I didn't want to diss OpenBSD, but it is hard to write a post about BSD without getting flamed/lectured/both by someone.

Re:No SMP? Huh?

[Korpo]

While I largely agree with your reasoning, I have a question:

Wasn't it OpenBSD that (on purpose) broke its IPv6 subsystem in a release not long ago? This doesn't seem to relate well to "You know that the packet filter will play nice with the IPv6 subsystem".

I don't know IIRC, so don't flame me if not. ;)

Re: No SMP? Huh?

[gidds]

Erm, you are aware that Windows 2000 was effectively NT 5? And that Windows XP is effectively NT 5.1?

OpenBSD needs SMP

[Sloppy]

I'm glad they're doing this. I know OpenBSD's focus isn't performance, but the day is coming when all personal computers will be multiprocessor. Not because people will actually put two chips on all their motherboards, but because of dual-core chips. Dual-core is such a great idea that pays off with so much performance per area-of-silicon, that eventually [speculation follows] economy-of-scale is going to make single-core chips an endangered species. Maybe you'll still be able to get a single-core x86 chip from Via or Transmeta or someone, but I bet AMD and Intel will stop making them altogether.

So OpenBSD faces a situation where people are going to be running their OS on SMP-capable hardware, even when the user may not necessarily highly value SMP. The user will have SMP simply because the $100 chip he bought, is dual core. All of OpenBSD's competitors will be able to take advantage of this, so OpenBSD's performance would be so far behind, that even if performance isn't the focus, it'll look just plain wasteful and backward.

All developers who makes OSes that run on commodity hardware, need to wake up and deal with SMP, because in a few years, everyone is going to want it, rather than just the speed freaks.

Re:SMP didn't make it more secure, so not a priori

[Alan Hicks]

SMP is a scalability issue

SMP is also a security issue, a negative one. Implimenting secure SMP is not an easy thing to do IIUC. Making sure that processes access memory in a secure way on a multi processor machine is difficult.

Documentation

[badriram]

Might I add that they have probably the best documentation ever. It is extremely well laid out, and easy to follow. And useful, unlike man pages....

Yes, you are quite wrong.

[Anonymous Coward]

SMP webservers are a waste of money. HTTP is rediculously easy to cluster, so it makes more sense to buy a half a rack of cheap single CPU machines, giving you more redundancy, and better performance than large SMP servers. Typically cheaper this way too. This is pretty much why linux is hurting sun so bad.

Um, not quite.

[Anonymous Coward]

There were a couple problems pointed out by those benchmarks, which were fixed shortly after that. More so, the test was flawed and some of those results are just plain wrong. I have benchmarked the latest openbsd, freebsd, netbsd and mandrake linux, (using actual benchmarks, apache, sendmail, etc), and freebsd 5 is noticably slower than the other 3, which are all pretty much the same on single CPU machines.

Too little, too late.

[Anonymous Coward]

OpenBSD has dragged its feet far too long in adding SMP support, that I'm afraid it's fallen off the radar screens of far too many would-be users, who now know that they can make a Linux or FreeBSD machine just as secure, plus get a lot better performance and be much closer to "mainstream". OpenBSD once had its famous claim to "no remote exploits since blah blah blah" but even that doesn't bear the weight it once did in the past. IMHO, a few years ago OpenBSD had its 15 minutes of fame, and now has fallen into irrelevance. Had SMP support been implemented parallel to the same time or shortly after FreeBSD was ironing out the kinks, OpenBSD would not have lost so much popularity-momentum. I ran it as my company's Internet firewall & server platforms until the uniprocessor boxes couldn't keep up anymore and since my replacement hardware was dual cpu, I just made the switch to Linux instead since the new hardware was much better supported, and have had _ZERO_ security problem with Linux on these boxes since then, just by keeping them configured and updated properly. OpenBSD has no security edge or capabilities over Linux nowadays. It's only just a quaint little academic curiosity anymore.

BeOS had SMP back in 1995!!!!

[Anonymous Coward]

When the BeBox was announced way back in 1995, BeOS already had support for the dual PowerPC cpus in the machine!!!

Re:No SMP? Huh?

[jazman_777]

Uhh, am I the only one here who was completely unaware that OpenBSD didn't have SMP support?

I guess I've been stuck in the Linux world too long. I never would have guessed this. I haven't been running a single-processor system, workstation or server, since 1998.

The bigger news here, for me, is that Linux just jumped way up on my totem pole of respect.

So, you're completely ignorant of what goes on in the OpenBSD world (hint: security); by your own confession you've been "stuck in the Linux world"; and now you respect Linux more? I guess I can see how you might think the OpenBSD guys are a bunch of non-SMP doofuses.

Re:No SMP? Huh?

[jazman_777]

The OpenBSD pf has security features that cost large sums of money in the closed source world.

Indeed, and don't forget the firewall failover capability. It's stuff like that makes the security folks swoon.

Re:No SMP? Huh?

[ndunn]

I had mod-points, but I couldn't find "blasphemer".

Congratulations!!!

[Anonymous Coward]

I knew you guys could do it! Welcome to 1995!

Re:No SMP? Huh?

[dasunt]

IPv6 is irrelevent right now outside of lab environments and isolated networks (cell phone providers, etc.).

How about this then: OpenBSD had ipsec in its core long before linux ever did. The vanilla linux kernel didn't get ipsec until the 2.6 series, everything before that was either 3rd-party kernel patches and daemons.

Hell, windows had ipsec before vanilla linux did. :(

Re:No SMP? Huh?

[Santana]

"Nearly every "serious" web server is SMP - at least in corporations, or am I wrong?"

You are wrong. You can have a serious web server without SMP. Serving pages is not a big deal. You need memory and fast disks, that's all. Unless you have Oracle + Java on it too, but that doesn't make the rest of the web servers less "serious".

Re:No SMP? Huh?

[SillyNickName4me]

Hmm, that may be true for Google since they generally don't have to keep state. When you have to keep track of sessions and need a substantial amount of sharing between your server instances as a result, a cluster of dual cpu machines or even quad cpu machines may turn out to perform a lot better.

At any rate, if your goal is to handle lots and lots of relatively independent queries, then yeah, large clusters of relatively small machines do a very good job.

Re:No SMP? Huh?

[Mr. Piddle]

"SMP technology" proved its worth decades ago.

Yet it only became popular in the last decade. Remember, Sun was a *workstation* company before the mid-1990s. The big-ass UNIX servers we see today really are a fairly recent development--before them were the mainframes. I think it is fair to say that SMP really is only recently proven, regardless of how long various implementations existed.

woot

[ForsakenRegex]

I've been waiting a long time to try OpenBSD on my SMP boxes. Time to see what happens.

Re:No SMP? Huh?

[Anonymous Coward]

Not to mention that true serious implementations involve multiple servers, not necessarily multiple procs. All else being equal, for most implementations i'd rather have 8 1-way web servers (w/ rr dns and carp or similar) than 1 8-way web server.

Netcraft confirms that QNX...

[CaptainPinko]

must suck. They seem to be hosting their site on Linux. Wow, now that word is out I doubt anyone will ever use them them again!!! QNX must be dying...

Your spoon will always make for a lousy fork.

Re:Netcraft confirms that QNX...

[ArbitraryConstant]

Netcraft runs FreeBSD...

Re:SMP didn't make it more secure, so not a priori

[Santana]

I wonder how secure FreeBSD's and Linux's SMP implemention are

Re:No SMP? Huh?

[the morgawr]

You can view the CVS repository and decide for yourself.

Right now I think most of the kernel is under big lock. SMP is currenly on target for a stable release in OpenBSD 3.6 (Nov 1, 2004). I suspect that by 3.7 (May 1, 2005) big lock will have been pushed down into at least the major subsystems.

It may happen even before then, the Calgary hack-a-thon is comming up and SMP is a major focus. Mostly they are working on bug fixes but I wouldn't be surprised to see a bunch of SMP improvments come out of that.

Also, if you read the tech@ list archives you'll notice that they are asking people to test the bsd.mp kernel and report errors. They are also looking to borrow computers that have certain types of hardware problems that are causing software issues (so that SMP will be supported in them).

Re:No SMP? Huh?

[the morgawr]

Also, SMP is a pervasive, system-wide thing, you can't borrow it from another OS, especially one so internally different as FreeBSD. (You couldn't even borrow SMP code from the fairly similar NetBSD).

Re:No SMP? Huh?

[ahodgson]

Dual CPU Webservers are generally considered bad, as they don't scale near as well as 2+ webservers serving files via nfs or other technology.

While that may be true in your case, I disagree for the general case.

When rack space costs you money, a dual-CPU 1 RU box cannot be beat for web-serving efficiency, especially if you serve any volume of dynamic content. For my $day_job, dual-CPU boxes can serve almost exactly 200% of the traffic of a single CPU box, and cost us no more in rack space or network ports or power provisioning than the single CPU box. Oh, and they require the same amount of sysadmin time too, not that that amounts to much in a web cluster.

Re:No SMP? Huh?

[the morgawr]

As I understand it, the NetBSD SMP support, like OpenBSD's, is available only in a developement branch (that will become NetBSD 2.0). Someone correct me if I'm wrong.

Re:No SMP? Huh?

[Anonymous Coward]

I wonder where Windows NT is on your totem pole, because it had SMP support years before Linux 2.0. And ACPI support and journalizing filesystem support and modules (drivers) support and so on.

Since I have yet to see a version of NT in production that can actually do more than one thing at a time without crashing, that's not so impressive ..

Re:No SMP? Huh?

[dekeji]

Yet it only became popular in the last decade. Remember, Sun was a *workstation* company before the mid-1990s. The big-ass UNIX servers we see today really are a fairly recent development--before them were the mainframes. I think it is fair to say that SMP really is only recently proven, regardless of how long various implementations existed.

Mainframes had symmetric multiprocessing going back to the 1960's (!).

And "popularity" is not a useful concept for measuring when something has succeeded in the computer world: because of the spectacular growth of the computer industry, almost any feature you can name has become popular only in the last decade.

However, in terms of fraction of total installed machines, I suspect there may have been more SMP machines in the 1960's than there are today.

Re:No SMP? Huh?

[Tranzig]

The blasphemer? He was running that way! *points to random direction*

Re:No SMP? Huh?

[ArbitraryConstant]

Linux is targeted to be a general purpose OS. Those get SMP support fast because it's important for various things (esp dynamic content, database, development, etc).

The developers concentrate elsewhere, and that's not an empty excuse, look at the results. ProPolice everywhere (Linux doesn't have this even though there's a lot of places where it would be easy), W^X (even though OpenBSD has shown that it can be done without major breakage), transparent firewall failover (Linux can do failover with UCARP, ported from OpenBSD, but any existing sessions are toast), other tweaks like ACK packet prioritization (which NetFilter can do, but it's a PITA and requires either blind copying and pasting or deep knowledge of TCP).

Linux is way better as a general purpose OS, no one denies that. But, I'd rather have two OpenBSD firewalls ready to take over for each other transparently than one Linux firewall. No matter how many CPUs it has.

Re:No SMP? Huh?

[fimbulvetr]

Though your comments are well founded, I still somewhat disagree. After seeing both comments, I can see cpu being useful especially if you are doing SSL connections and session tracking.

You might actually be one of the very few people who find that (double the cpu=double the power), but that's not very common for most web sites. Most of the time, dual cpus do not mean twice the power. I'd venture to guess that you might see 1.25 or 1.30x the power.

I intended my main point to be that web serving relies more on I/O and memory than CPUs.

My day job is an admin at an ISP and I find thousands of sites (ssl and reg) can run perfect on a 440Mhz Sun Netra t1. I have two of the $300 single proc bad boys runnning thousands of sites without a glich. When one fails, which niether have (Knock on wood), the other one is an easy failover script away.

When these finally hit 40% consistent load(current load = .21 and .7), I pop on a 3rd linux box.

And yes, they are very fast and incredibly stable.(as long as you keep solaris patched.)

Re:Time to move openbsd.org to OpenBSD then ...

[cpeterso]

but OpenBSD still looks bad for not "eating their own dogfood" on openbsd.org. What is Microsoft ran microsoft.com on Solaris servers? Remember the bad press Microsoft when they still ran hotmail.com on FreeBSD servers?

Not happy with features

[mnmn]

The strength of BSD over linux has been stability, maturity and predictibility. Linux has more features than any OS out there, and BSD has been a humbler, older, more serious OS with its developers looking less at the market and more at the buglist.

Of the BSDs, OpenBSD has been the most BSD of them all. FreeBSD is itself a bit of a Linux, slackware to be precise. It has enough 'features' to be usable on most hardware, highly ported, yet strong and reliable too. OpenBSD however has been the simplest, aiming for cleaner code, more security and the ultimate in stability. Thats the whole appeal of OpenBSD, not to run it on a dreamcast or playstation, or to run it on a 16-CPU unisys, or to make a TiVO out of it. Its as simple as modern UNIX can be. Its nice to have at least one OS out there where developers spend more than 90% of the time cleaning things up rather than adding mess you dont need.

So I'm not sure if adding the SMP is all that much of a great idea. SMP is tough as Linux can testify with its years of threading library woes and interrupt locks. Few OSes have SMP really pinned down well, Linux only realy stabilized that, and OpenBSD is now getting involved. I hope they start out with no threading, only full processes per CPU. We dont need performance or portability or scalability out of OpenBSD, there are other BSDs for that. We need it to keep its minix-like simplicity. OK bad example. UNIX-like simplicity.

Re:Not happy with features

[Anonymous Coward]

"We need it to keep its minix-like simplicity. OK bad example. UNIX-like simplicity."

UNIX-like simplicity? *sound of me banging head agains desk*

Quote:
"...There are too many flavors and variables to answer this accurately. Suffice it to say that virtually every Unix supports SMP."

Re:No SMP? Huh?

[phoenix_rizzen]

The first line of the second paragraph should read:
The inefficiency in the SMP in FreeBSD 4 is due to the fact that ...

In case anyone is wondering:

[drinkypoo]

40% Interesting 30% Informative 30% Overrated

Looks like someone deliberately waited until the initial flurry of moderation had ended and then slapped me down with the ol' overrated.

Overrated is the final refuge of the incompetent moderator...

Re:No SMP? Huh?

[Brandybuck]

I haven't been running a single-processor system, workstation or server, since 1998.

You, sir, are part of the great clean scrubbed minority. Most people do not have SMP systems, and probably won't for another five to ten years. I do not have one, and have not ever had one. None of the workstations at my work are SMP.

The bigger news here, for me, is that Linux just jumped way up on my totem pole of respect.

This is just OpenBSD we're talking about here. Other BSDs have had SMP for quite some time. Did this news also bump FreeBSD, NetBSD, DragonFly and Darwin up on your totempole?

Re:No SMP? Huh?

[Brandybuck]

BSD had IPv6 before anyone else. But that shouldn't surprise anyone, because BSD had IPv4 before anyone else did either :-)

Re:Time to move openbsd.org to OpenBSD then ...

[Brandybuck]

Remember the bad press Microsoft when they still ran hotmail.com on FreeBSD servers?

There are very strong rumours that they still do on the back end where no one on the outside can see them.

Re:No SMP? Huh?

[macdaddy]

Ok, we're not talking rocket science here. Basically what OpenBSD did today is add electronic fuel injection to their "secure" car. They still don't have an anti-lock brake system and are still using a straight pan-head but they do have the best damn seat belts and air bags in the business. Too bad their utter lack of attention to common performance enhancment technologies means you'll never go fast enough to NEED those super duper safety features.

Comment removed

[account_deleted]

Comment removed based on user account deletion

Re:Time to move openbsd.org to OpenBSD then ...

[murgee]

There was a page on this. I can't seem to find it anymore, though.

OpenBSD doesn't own openbsd.org - it's run by the U of Alberta SunSITE, which is one of these things where Sun gives your university a bunch of hardware. There's a whole bunch of 'em around the world (there's a list at the link). They're kinda precluded from running anything besides Solaris on it.

Re:No SMP? Huh?

[Anonymous Coward]

You are an idiot: IPv6 support is a mandatory requirement to have as IPv6 rollout comes ahead in the next few years. It's a testimony to the solid reliability of the BSD protocol suite that the main experimental development for IPv6 (notably via. KAME) was undertaken on BSD.

BSD is fucking dead whether you want to believe it or not. Take all the installations of *BSD, add them together, and multiply them by 10 and they STILL number less than the number of Red Hat Linux installations for christ's sake!

Re:No SMP? Huh?

[Anonymous Coward]

OpenBSD had ipsec in its core long before linux ever did. The vanilla linux kernel didn't get ipsec until the 2.6 series, everything before that was either 3rd-party kernel patches and daemons.

ipsec sucks even worse than ipv6. I wouldn't allow that shit through my firewall. It's like giving the inmates the keys to the asylum letting them run whatever the fuck they want through a tunnel in your firewall.

Re:No SMP? Huh?

[nacturation]

They still don't have an anti-lock brake system and are still using a straight pan-head but they do have the best damn seat belts and air bags in the business. Too bad their utter lack of attention to common performance enhancment technologies means you'll never go fast enough to NEED those super duper safety features.

What? Since when is system security a function of performance? A better analogy might be having a Ferrari and a Honda. Doesn't matter how fast it goes, if you come under small arms fire you're just as dead. Or your computer is just as hacked. Adding bulletproof glass might slow the car down due to extra weight or something -- yeah, stretching the analogy real thin here. But if security is a concern to you as a sysadmin, would you rather have an insecure fast system or one that's potentially a bit slower but resistant to attacks?

Re:No SMP? Huh?

[Anonymous Coward]

Do you want the hash or the weed?

OpenBSD HEAD: Obligatory Movie Quote

[Anonymous Coward]

HEAD! PANTS! NOW!

Re:No SMP? Huh?

[lsdino]

I think in general you point to one of the greatest weaknesses in Linux. For all the bitching you hear on Slashdot regarding X-Windows, or Gnome spatial interfaces, or KDE being slow and bloated, the biggest problem is QA.

It's not really the fault of any individual team. They're going to write their code on Linux distribution X, and test it there, and if you're lucky they'll run less grueling tests elsewhere. Ultimately you'll get to the distribution level where you'd hope everything gets tested once and for all. But does Red Hat pick up Ximian's test suites? Or do they just build, run smoke tests, and ship? Certainly they'll test the kernel, as they have active developers there. What about all the other people's source they ship?

I think there may be a bit of blind cheer that integration issues somehow don't exist in Linux land. That components act over clean interfaces and don't take any reliance upon implementation issues. But this is a problem that software engineering has faced since the dawn of the industry, and I don't see a smoking gun pointing to how Linux has solved this problem. And until this problem has been solved the best thing our industry has to offer in it's abscence is end to end testing.

What Linux does have going for it's self is hordes of developers... But lots of cooks does not make a good cookie.

Re:Time to move openbsd.org to OpenBSD then ...

[evilviper]

What is Microsoft ran microsoft.com on Solaris servers?

Microsoft has control of their servers that run Microsoft.com. OpenBSD does not have control of the server behind OpenBSD.org, since it is at the University of Alberta, and being used to host other project as well.

You'd be just as well off complaining that RedHat doesn't eat "their own dogfood" because their ISOs are distributed from many mirror sites that don't run Redhat Linux.

Re:No SMP? Huh?

[evilviper]

windows NT rocks. Its the only version of windows i had any respect for

Please shut up.

"windows NT" does not mean NT4.0. In fact, in this discussion, it actually means NT3.51, which was the first to have multiple processor support, IIRC.

That said, I'm a fan of NT4 myself. If I must use Windows, I'd much rather have the simple, stable, and very fast NT4. USB support and DirectX >6 be dammed...

Re:BeOS had SMP back in 1995!!!!

[Anonymous Coward]

Why don't you log in, Eugenia?

Re:No SMP? Huh?

[Shanep]

The OpenBSD project simply has too few resources, and that's for sure! And that has nothing to do with "different set of priorities then Linux". I'm pretty sure, that Linux developers are numerous enough to aim for all three goals: reliability, safety and efficiency.

SMP opens up huge security implications, it has absolutely everything to do with "different set of priorities then Linux".

OpenBSD has a better security track record than Linux, with those numerous Linux developers and manages to deliver it without hurting functionality.

I think a big reason OpenBSD might be pushing forward with SMP now, is due to multicore CPU's and technoligies like HT.

Before that, with security being the prime focus of OpenBSD, they would have been silly to put any efforts into SMP, since the performance gains are nowhere near worth the security hassles.

Just in Time

[Anonymous Coward]

Since it appears that in the next few years the only way to get a single core computer will be to use an embedded design, as it looks like all the regular desktop/server chips are moving to be dual core.

Re:No SMP? Huh?

[Anonymous Coward]

Dude,

I'd bet you're "lack of scalability" comment stems from Felix's little meaningless benchmark suite. You really need to stop looking at other people's pretty little graphs and learn how computers work.

Re:No SMP? Huh?

[Morth]

That's why you run it in transport mode and then use ipip for the actual tunnel. Last I looked Linux didn't have ipip either, but I think it was planned for 2.6.

Sure, it would be much better if the ipsec traffic came on some pseudointerface, like it did/does in FreeSWAN, completely agree there...

Re:No SMP? Huh?

[TheRaven64]

Bah! You and your `facts' and `accuracy'. Don't you know this is Slashdot?

Re:No SMP? Huh?

[pairo]

Whoaaaa. Let's not go there. From net/ipv4/ipip.c: Authors: Sam Lantinga (slouken@cs.ucdavis.edu) 02/01/95

Re:No SMP? Huh?

[Morth]

Figures... didn't think it was there because I couldn't find any way to activate it in some 2.4.x kernel.

Re:No SMP? Huh?

[mirabilos]

Of course.

I won't take it over from OpenBSD, either.

There are various reasons for this decision,
for example "either correctly or not",
security and "doing it would mean a rewrite of
90% of the non-driver code (and a good bunch
of driver code), and that won't be real BSD
any more".

my HEAD multi-tasks too

[MMHere]

I can do several things at once. I guess I've had SMP in my HEAD for some time now.

Re:No SMP? Huh?

[macdaddy]

Bad analogy. I'm a compotent sysadmin. I can make any system secure enough to meet my needs. Yes, even Windows. I don't need O-BSD folks second-guessing my abilities as a sysadmin.

Want some more of my karma?

[drinkypoo]

20% Interesting
50% Overrated
30% Informative

Hooray for abuse of moderation! Welcome to slashdot, can I take your coat please?

Re:No SMP? Huh?

[phoenix_rizzen]

You can be as innacurate as you want with any post relating to Linux or Windows. When it comes to FreeBSD, though, it better be 100% correct. :) Gotta keep the balance, you know. :D