Forgot your password?
typodupeerror
BSD Operating Systems Hardware

NetBSD/Alpha goes multiprocessor 87

Posted by Nik
from the one-banana,-two-banana... dept.
chaoskitty writes "Jason Thorpe has gotten multiprocessor NetBSD/Alpha to go multiuser! He has already done a kernel build and full userland build with the multiprocessor kernel. More details are in Jason's message to the NetBSD tech-smp list.
This discussion has been archived. No new comments can be posted.

NetBSD/Alpha goes multiprocessor

Comments Filter:
  • Otherwise I'd be severely disappointed, but I'm sure you meant multiprocessor. Now we'll just wait and see how many posts we get on "Wow, it just went multiuser when Linux has been doing this forever?? No wonder *BSD is dying!" ;-)
  • now if I can get my SMP athlon out of amd.
  • i like the fact that there is no closing bracket to the italix...it makes the page look a little different. but..enuff complaining. oh...and so this is ontopic....BSD is cool except i don't like writing my own drivers.

  • I have to say, the alpha part of the post kinda slipped out of the content when I was reading it... First post anyway...
  • No, the multiprocessor kernel is now multiuser.

    ---

  • It will happen. And it will be sweet.

    --Lenny
  • No, the news is that the multiprocessor is now multiuser.

    To quote: "I got multiprocessor Alpha kernels to go multi-user, running both user and kernel code on multiple processors".

  • by bartyboy (99076) on Sunday April 22, 2001 @05:48PM (#273520)
    If anything's an indicator that BSD isn't dying, check the number of BSD articles posted today on /.

    Then again, they could be hitting a niche market.

    (awaiting troll moderation)
  • actually he does. multiuser is the state where unix like os's allow one to log on as more than root... ie, their normal state of operation. in sVr4ish systems (like linux) this corresponds to runlevels above 0.
  • uhm, it was about smp on alpha.

    I'm pretty sure the bsd's (except for openbsd, maybe) have had smp as long as linux or even longer, if you count being stable..

    --

  • Hah! Close your tags next time!
  • I know this is probably a really stupid question, but I'm curious:

    Do you have to rebuild your applications to take advantage of this? Will binary distributions still take advantage of the SMP speed boost, or do you need to recompile them with SMP support?

    I know with Linux it's a mixed story, just wondering if it's the same deal with BSD...

    beat me with clue stick, please.
  • amdsource.com [amdsource.com] have an article in their news section that talks about dual Athlon MoBo's being beta-tested.

    Looks like they're coming soon to a puter shop near you!
  • Goddamn it, Gay Nik, close the fucking italics tag. The whole front page is on a slant.

    --

  • I don't follow BSD too closely, so I'm somewhat curious (if someone doesn't mind indulging my ignorance): I'm getting the impression from the other posts that, up to this point, the x86 architecture was the only one that supported SMP under BSD. Is that how it stands with BSD (plus the Alpha architecture now), or are there other architechtures it supports SMP on?
  • NetBSD != BSD in general.

    FreeBSD has SMP support [freebsd.org] and I'm pretty sure various BSDs have had SMP support on various architectures other than the Alpha.

  • Is a great step forward for NetBSD/Alpha.
    I've used NetBSD on a wide variety of platforms, I can definately say that its been a lot more usable than other OS'es I've tried. Yes, its not a big deal in comparason to Tru64, which can run on a disgusting amount of CPU's on Alpha Hardware, but hey, its a start.
  • The article states NetBSD/Alpha, not NetBSD/x86...
    Your MP Athlon will be nice, but it won't be useful for NetBSD =)

  • Have you ever seen linux actually running with 32 Processors? You'd be suprised as to how many people haven't run any OS with 2 CPUS... Personally, I've had some very horrible experiences with Linux/SMP, in fact total disasters. After switching to FreeBSD and SolarisX86 on the same boxes, I got better performance and no more issues. Personally, I think its neat that linux can run on a buncha cpu's, but the fact that it does it doesn't mean you should use it...
  • Any app which either uses system threads, or forks child processes, will automatically be split amoungst processors at the behest of the task scheduler. Doesn't require a recompile, unless there's green threading vs native threading or some such, but it does require the app itself be written with some form of multiprocessing in mind. Such enhancements also have benefits on single-processor machines, so quite a few are so written these days, where appropriate.
  • I believe the article implies that the machine went multiuser on a multiprocessor machine.
  • Or you could use FreeBSD or BSD/OS.
    Both are Very BSD and they both run on X86 boxes with more than One CPU (and can use the other CPU's)
  • Since when is FreeBSD an SMP-capable platform? Have I been living under a rock?

  • He did mean multiuser. Before now, NetBSD only ran on multiprocessor Alphas in single user mode, for whatever reason. Now that it runs in multiuser mode, it's on par with the other NetBSD ports, although from the web page it doesn't support all multiprocessor Alpha boxes out there, yet.
  • He *did* mean multiuser. There are quite nasty issues involved with getting SMP systems to run both user and kernel code on multiple processors. It's non-trivial. Go read some of Modern Operating Systems [amazon.co.uk] by Andrew Tanenbaum for a good explanation - it's much too complicated for me to summarise concisely in a /. post.
  • Just when I thought I'd seen the ultimate bsd troll, see that it's spawned yet another.
  • OK. I'll have to bite on that one. JoeLinux user shouldn't open his mouth to flame until doing his homework. FreeBSD has had SMP for eons....
  • Negative on the BSD x86 SMP only. MacOS X supports SMP on PowerPC G3 and G4 dual CPU systems, and it's BSD all the way.
  • Have a bunch of cool things happen to *BSD, we get to flood slashdot with anouncements. Have one very other day, some go unnoticed. Density is the key! Save 'em up and release them at the end of the week.
    <joke>

    somewhat true, somewhat funny

    ---
  • Have you ever seen linux actually running with 32 Processors?

    I can't claim 32, but I do have more than a little experience in the 4 to 8 range...and the 2.2 Linux kernels don't use more than 4 processors all that well. Testing real-world stuff that does more than just crunch cycles didn't show a lot of improvement when going from 4 to 8 processors in the same box (a Compaq ProLiant 8500 with 8 GB RAM, only 2 GB of which was used in testing).


    Even so, the lack of SMP support in the BSDs (and no, One Big Kernel Lock is *not* SMP support) has been a real drawback, so I'm happy to see this happen.
    --

  • Ok, some clarity seems to be needed here.

    1) This is news.
    2) Alpha is a type of CPU. DEC made 'em and since spending US$9.5 Million on a Monday in 1998, Compaq make 'em.
    3) NetBSD runs on many platforms
    4) On this paricular platform (alpha), NetBSD has had Multiprocessor alpha for a little while, but it dosen't matter much until you can make it go multiuser.
    5) This is news, because compaq make some "big-ass" alpha systems (32x Alpha = Floating Point like AMD and Intel dream of :-)
    6) Our sweet lil OS (I run it on Alpha and VAX DEC equipment) runs on even more big systems, this may even be the biggest.

    If you have a spare alpha + vt320, then so the lil box a favor and see http://www.netbsd.org/Ports/alpha/ then join port-alpha@netbsd.org.

    It clensed my soul. :-)
  • I thought these were abandoned by Compaq.
  • What advantage is there to running NetBSD on Alpha vs. FreeBSD? I'm not intending to start a flame - I do believe diversity and choice is better...but I'm not sure I see why, with the proven robustness of FreeBSD, one would choose to run NetBSD?

    Is there technical merit to it? I'm just curious.

    sedawkgrep
  • Possibly :)
    Its been SMP capable since 3.0
    Sure it wasn't great then, but during the 3.X-RELEASE branch its been pretty damn solid on the systems I used it on. 4.X I saw a lot of improvements, and now with 5.X its gone through a redo to make it even faster than before. FreeBSD/SMP was already really fast with the limitations it had, now it'll be even faster.
  • I remember reading a while back(on linux-kernel I believe) of somebody getting Linux to run on a 24 cpu Sun E10000. I'm not sure how great the performance would be, but it is possible.
  • nah, whenever people come out with patch branches, you've got different versions. They don't get integrated in till El Dictator (Linus) decides the patches get in to the main kernel. Till then, they're separate, and essentially a different version.
  • by Alien54 (180860) on Sunday April 22, 2001 @07:53PM (#273549) Journal
    Okay, it's late, I'm bored. so here we go. As you said:

    My Reply
    It's about time someone posted a pro-MS article on this site - the only thing I'm dissapointed with is the number of troll statements posted in the comments.

    And then yo did an extensve analysis of the article.

    The problem is that the start of your comments were *cut off* well before the start of your commetary by the infamous line "Read the rest of this comment... ". In other words, what was visible was the most troll intensive portion of your article. This was very bad positioning for your comments, since some of them were mildly interesting. Probably such a post should be reorganized, and should have been saved for posting in reply to an appropriate article. Here it is best seen as a troll.

    Now there is the matter of this bit:

    Why would individuals encrypt their emails and other correspondence to each other? What is the rational explanation? The only reason I can see for day-to-day use of encryption is personal emails is that you have something to hide or you have a bad case of paranoia. No offence people - but what makes what you say so interesting that you are so concerned about other people reading it? If you are doing something illegal, or you are concerned about maintaining secrecy because other people may steal your original (and so far unpatented) ideas then maybe there is a point - but I have met some people who refuse to exchange email unless it is PGP encrypted - what's up with that?

    Simply, PGP offers a way to ensure that the person sending you the email is the person they say they are. Not some one faking it.

    The other issue is one of Privacy.

    If you do not belive in privacy, then I can recommned a glass house for you.

    After all, you are not doing anything illegal? And if all houses were made of glass we would be able to catch criminals alot easier. We could just watch them all of the time with TV cameras.

    What are you doing that is so important that it would require secrecy and privacy 24 hours a day? You must have a criminal frame of mind, not wanting to live in a glass house.

    This obsession with privacy is merely paranoia, y'know, and is easily fixed with one of several medications. Let us recommend a nice doctor who would be very willing to help you with medications.

    Check out the Vinny the Vampire [eplugz.com] comic strip

  • The first multiprocessor BSD [purdue.edu] was running in 1981, a decade or more before the first SMP Linux. (The reference mostly describes the hardware, which was an even better hack than the software.)
  • Otherwise I'd be severely disappointed, but I'm sure you meant multiprocessor.

    Both. Getting a multiprocessor kernel to boot is a pain. Getting it to do fsck and some other I/O is a pain. Getting it to run all the other userland boot code isn't so hard (once the I/O works). Going to multiuser mode isn't really hard, but then the network services start up and having anyone poke at them can be a chore. Having a user actually login and run crap is also another level of pain.

    I think past that is an actual release :-)

  • I run NetBSD on three architectures at home. Some of my i386 boxes, my Sparc boxes, and my 68K Macintosh boxes. The same OS runs on all three architectures. It all builds out of the same source tarballs. The configuration files in /etc are identical. The same apps will build on all the boxes out of the NetBSD Packages collection.

    There's a lot of merit in that for various reasons. I grew tired of every yahoo out there tweaking the config files for a certain other freenix and calling it a 'new distribution.' It gets old after awhile.

  • netbsd was on dream cast way b4 linux was. i remember readin about netbsd gettin on it a while back. then i saw some post about linux gettin on it and i was like "yeah... netbsd's been doin that a while."

    --
    Tres_Status
  • "Oh well, as long as BSD has some cool "3l33t d00d" image"

    Why do I see this every time a BSD/Linux argument comes up? Anyone who makes a choice based on how it will effect their image, or "leetness," is nothing but a lamb and should be shaved naked anyhow.

    "openbsd might be marginally more secure as long as you don't install anything besides the defualt software, but then it only runs on 1 x86 processor, so using it as a serious server is pretty much out of the question."

    I'm a big OpenBSD fan. Lack of SMP support is not going to drive me away from choosing OpenBSD for the majority of work that I do. There really is no reason that most servers would need multiple CPUs, unless the applications you run are constantly hitting over 80% or so usage on your current CPU.

    If I was in some situation that required SMP support, I would evaulate which OS provided the best support for it, as well as check up on any other qualities I might like to have in that OS which I choose to use. I have used both Linux and FreeBSD on light dual Pentium 3 machines, and really haven't noticed much difference between the two. So, in that case, I obviously stayed with FreeBSD because it is my personal preference.

    Honestly, I don't know where the argument is. Everyone seems to be arguing over how many CPUs which OS can run, and in the case of someone earlier stating that Linux could run on 32 CPUs, well, why are you all worried about such a thing when the majority of you will never own - or even see such a machine in your lifetime?
  • Where can i special order *BSD with 32 processor special hardware? Oh oops i can't.

    No special hardware required, apparently, with NetBsd, aside from the usual, like a computer. If you want proprietary hardware, you'll have to go to Microsoft, I guess

    Check out the Vinny the Vampire [eplugz.com] comic strip

  • Wow, I can assume you must be an excellent kernel hacker to be saying this with such authority. From the way you stated that with such a forceful manner I could tell that you have written code for both systems and know the SMP code inside and out.

    I bet you are subscribed to freebsd-current and freebsd-hackers and post to them often about how they could improve their code. With your vastly supierior knowledge of the way it is done. I know that you are only posting this because you have seen over and over again that your code patches, that would make FreeBSD (or any BSD for that matter) up to par with Linux, rejected by the FreeBSD team. And you are unwilling to split FreeBSD into another branch.

    I bow before your holiness. Please teach me how to become as kernel literate as you.

    Of course, in reality you probably couldn't tell SMP code from a device driver. So why don't you come back with actual knowledge of what you are talking about. Please note: as I am not deeply involved in either one of these kernel's SMP development I am not able to say which is better. Nor am I willing to make an ass of myself by trying.

    ---
    "Do not meddle in the affairs of sysadmins,

  • Sorry....I meant SMP on alpha...my bad...I apologize...

    JoeLinux
  • I was actually going to comment that... "So I was doing this on a DEC Alpha 3000/300L back in '95 running OSF/1".

    Not even sure Linux ran on an Alpha back then.
  • You troll, but it's the truth.

    I was once present in a bar in Cambridge where there was scheduled to be a BSD user's group meeting. I was not paying much attention as I was with a group of friends from work and we are all Windows developers. Besides, they looked scary. So about I would guess halfway through their meeting, in walks ESR a group of his Linux lackeys. Well naturally the BSD elitists begin teasing the Linux guys, who pretty much just blow it off. But then one of the Linux guys, who was a kernel developer, I forget his name, says something to the BSD devil chick. I didn't hear what because there was a live band and it was pretty loud, but you know it wasn't good. So one of the BSD guys (one of the scarier of the bunch.. and this was in the presence of ESR, and you know how scary he looks..) starts getting all up in the face of the Linux guy. Well the Linux guy poors his beer over the BSD guy's head. Well, shit. There were also some foreign grad students from MIT in there. So one of them, who is a Pokemon master, gives a Pokeball to one of the BSD guys. So he throws it down in the middle of the bar, and what do you know, it's a charmander. Now I've learned this the hard way but in case you didn't know: alcohol and fire do not mix well. So the charmander breathes fire all over the Linux guys, two of them are set on fire (they were working on USB support...that explains that) but the rest are fucking pissed now. So one of them takes a baseball bat of the wall and underhandedly swings at one of the BSD guy's legs. So he breaks his knees. This was a shame since he was an olympic swimmer. Too bad. So the BSD kernel guy got pushed over the edge, and starts beating the shit out of the Linux kernel guys. Finally he hits him so hard his teeth fly all over floor, and he begins kicking his head. He probably would have died except then ESR took out of gun and blew his head in 'twain right there.

    The lesson here is, it is a good idea to have ESR on your side in a fight, because while he looks scary but not that scary, he does carry a gun and is freakin' crazy.
    ---

  • Linux won't run on the 3000/300L now either. :)
    That pesky TURBOChannel bus.
    (The one on my desk is running OSF/1 at the moment)
  • You have to get them at one of those bulk discount warehouse clubs.
    ---
  • do some research there bud. FreeBSD has been SMP for some time. Maybe not the best implementation but it's quite useable. 5.0 promises some much needed improvement for FreeBSD's SMP.
  • The article states NetBSD/Alpha, not NetBSD/x86... Your MP Athlon will be nice, but it won't be useful for NetBSD

    But the SMP support will probably be relatively straightforward to port to x86, especially given that Alpha and recent Athlon processors share a bus protocol.

  • Well it may suck, but from Wind River's perspective it wasn't something they wanted to bank roll. You cannot blame them for that.

    Does it suck? As I said for slackware users it definitely does. Was it a surprise? No it was not.
  • How about LSDUserBSD

    That actually makes sense, given that both BSD and LSD were developed at UC Berkeley.

    --
    Why do dataBASEs have ACID properties?
  • by thorpej (316096) on Sunday April 22, 2001 @09:47PM (#273566)
    Actually, it is just what I meant. While uniprocessor kernels for NetBSD/alpha have run well for a number of years, I had only been able to get multiprocessor kernels into single-user mode previously. If you have never done low-level debugging of a multiprocessor capable kernel, then you probably don't know just how big of a milestone this is.
  • >>Negative on the BSD x86 SMP only. MacOS X supports SMP on PowerPC G3 and G4 dual CPU systems, >>and it's BSD all the way.

    MacOS X is a NeXT Mach kernel, derived from the CMU code, and that's where its MPU smarts come from, no BSD about it.

    The BSD part is the process accounting, device model, init configuration, userland and etc., above Mach. This has been implemented as what is called a "Mach Server" in the old CMU stuff.

    Jeremiah

  • NetBSD/i386 MP does work, but is not in the main source trunk yet -- it's on a development branch, waiting for some things to be ready for prime-time before it's merged. MP support for NetBSD/i386 will be in the NetBSD 1.6 release. The i386 port had a bit more of a challenge than I did on the Alpha port -- interrupt handling for MP systems is *totally* different than UP systems on the IA-32.
  • Don't you know porting NetBSD to video game consoles and scientific calculators and other crap will help it make many inroads with major corporations and increase it's legitimacy as a enterprise ready operating system...

    Actually, it does help make such inroads, but it's as an embedded OS, not an "enterprise" OS. Even the NetBSD folks will admit they aren't really shooting for the latter. But porting to "video game consoles and scientific calculators" proves how versatile NetBSD is as an embedded platform.

    Alphas are used in high-end embedded systems such as those used for industrial and medical imaging, since they still beat Pentiums two-to-one or better in the computations involved in such tasks. And to reach the useful memory and I/O bandwidth of recent Alpha systems, you have to use exotic chipsets with Pentiums, which can offset some or all of their price advantage.

    Wind River's purchase of BSDi suggests an important trend for the BSD's into the embedded market, a market that is expected to exceed the enterprise market in value in the years to come. Given Linux's problems in this area (lack of an integrated userland and what is often perceived as a corporation-unfriendly license) the BSD's have a pretty good shot at becoming open-source solutions for embedded systems.

    -Ed
  • Well, while the code hasn't made it into any releases yet, NetBSD supports SMP in some form on x86 [netbsd.org], alpha [netbsd.org], macppc [netbsd.org], and sparc [netbsd.org]. Perhaps others too...
  • This looks a lot like a shell to me... you can even telnet into a Dreamcast. http://www.netbsd.org/Ports/dreamcast/dmesg/010202 .txt [netbsd.org]
  • by tshak (173364)
    Where can i special order *BSD with 32 processor special hardware? Oh oops i can't.

    Oh, oops, clustering is more cost effective.

    Disclaimer: I personally don't like clustering databases, so your point has some validity to it :)
  • Actually, it can be very different in a variety of ways. In some cases, API's changed, which can be a really big deal. If its 'just a kernel update' then why would an API change drastically? Granted, I am not one programming with Linux Kernel API's at the moment (I happen to enjoy my BSD systems), but I would be pretty pissed off if an API broke in the middle of a 'just a bunch of patches' update...
  • My thoughts exactly... Compaq killed NT Alpha in late 1999, but it definitely was cool. (Even though most UNIX guys choose to ignore its existence, because the thought of an object-oriented OS with a decent GUI running on Real Computers gives them cold sweats.) It was a testament to the fact that Redmond had overcome its DOS roots (Or so we thought -- I don't understand how Redmond can live with itself after selling 95, 98, and now ME. Well, I guess the billions of dollars help.) and was capable of producing a portable OS -- an OS that, I might add, would be running on PPC and SPARC if it weren't for the RISC/UNIX Conspiracy Factor. (McNeally keeps the NT/SPARC license under his pillow, and dry-humps it, giddily, when he's in the mood... for love.)

    So while NT won't replace UNIX anytime soon, companies everywhere are discovering what the RISC/UNIX Conspiracy doesn't want you to know: it may take several dozen NT/IA boxes to equal the power of one enterprise-level UNIX/RISC box, but the Intel hardware is so fucking cheap that you end up saving money anyway. Even when you add Microsoft Hidden Costs ("$800 per server license? Are you out of your Vulcan mind, Spock?"), you pay less. And because IA is IA, you'll pay thousands of dollars for memory upgrades, as opposed for hundreds of thousands. (Last week I was watching our IT guys add some RS/6000 boxes to our cluster, and they let me hold $400k -- holy shit, $400k -- worth of RAM. It's a funny feeling, realizing that you could run for the door, sell the RAM, and buy a house with the profit.)

    Anyhoo, I think that the NetBSD freaks should stop trying to port to dying architectures and instead implement SMP on a certain lamer NetBSD-ripoff, [openbsd.org] whose lack of SMP is the only thing keeping it from attaining a level of mediocrity comparable to say, Cheese-Wiz. But hey, these are the guys that port software to fucking toasters for the hell of it, so we can't expact them to contribute anything valuable to the world.

    --

  • Well, as someone who's used both, I'd have to say go with which ever you're more familiar with. I believe they'll run binaries from either system on eachother now. FreeBSD has a *massive* ports collectio, while NetBSD's is not so big. FreeBSD has your nice friendly installer that you would be familiar with on the X86 side. Source trees for FreeBSD are the same on both ends, what hardware support FreeBSD offers for X86 also goes to Alpha (for the most part). NetBSD now supports Alpha-SMP, so if you have a box with more than one CPU, there you go. However FreeBSD apparently has this coming along as well. I like FreeBSD personally, but NetBSD is a great OS too. I've run it on my Alphas, Sparcs, X86's and Macs... FreeBSD only supports x86 and Alpha, so if you only have those two platforms, there ya go! If you want more than one hardware platform, but the same OS, NetBSD...
  • Ask someone who's worked with both, and they will tell you that Solaris has superior SMP at this point in time. Simply put, I converted a website from Linux based dynamic content servers to Solaris X86 and the site was about 4x faster just by doing that. Explain that? I had several people review my configs, they were fine according to them. I had *SEVERAL* problems while using Linux, so converting to SolX86 worked great. We worked those boxes hard, so thats possibly why you're having no issues, no serious load.
  • I didn't say I was running FreeBSD on my big SMP boxes, I was running SolarisX86. Pay attention AC...
  • >(and no, One Big Kernel Lock is *not*
    >SMP support)

    What a silly thing to claim, of course it is. In your opinion, how many "spinlocks" does the system have to use to have SMP-support? More than 1? More than 2? Is your limit the same for tiny message passing kernels as well?

    Regards, Tommy
  • For my own personal edification, I did a SMP shootout with my company's hardware (fun fun!).
    I had a 4CPU Compaq, and an 8CPU Compaq, each box had 4GB of ram. Linux couldn't address all of it, so it only ran with 2GB, I did find a patch that supposedly made it work with 4GB, but It didn't really seem to make a difference. Both the FreeBSD and Solaris systems supported 4GB, and probably would've used more had I tossed more ram into the machines. Those lovely removeable drives made it *super* easy to do comparasons between systems. I loaded up Linux, FreeBSD, and Solaris X86. I have to admit, I didn't particularly get great SMP support out of FreeBSD, but it did work fine, was relatively quick, was stable, and yes could handle more load. Solaris X86 was *the* hands down winner. I got the boxes to finish load tests 2-3 times faster with SolX86 - No crashes, no burps. Linux crashed on me 5 times during the 6 hours testing period, so I didn't get really good results out of it. FreeBSD didn't do anything bad, nice solid stable performance. However, I did manage to get the Linux system to do some work for a continuous period of time before I finished up the test on it, however it corrupted ext2fs and I lost the data (whee). From what I saw of it before the data went *poof* the Linux box was marginally faster than the FreeBSD system (yes, even with the big giant lock). If FreeBSD's SMP support is that fast in the state its in now, I'll be very pleasantly treated to even better performance.
  • I've worked with both... extensively. I'll say this much: I've never used Linux on a machine with more than 2 CPUs, but on a 2 CPU machine a 2.2 kernel runs very fast. I don't believe that Solaris X86 would be faster _period_ let alone 4x faster on a 2 CPU box. But I have used Solaris on 4+ CPU Sun machines and such machines scale very nicely. As of the 2.2 kernel, I'd say that on a box with 4 or more CPUs Solaris probably works better, and from what I've heard the gap grows wider as CPUs are added. Nevertheless, on a 2 CPU machine Linux is faster. I've not yet played with a 2.4 kernel on a 4 cpu machine.
  • by Arker (91948)

    You'd be suprised as to how many people haven't run any OS with 2 CPUS

    Quite true. The vast majority of boxes are still uniprocessor, and probably will be for awhile.

    Personally, I've had some very horrible experiences with Linux/SMP, in fact total disasters. After switching to FreeBSD and SolarisX86 on the same boxes, I got better performance and no more issues.

    FreeBSD and Slowaris both have had solid multiprocessing support for awhile. Sun of course makes their biggest margins on the boxes with lots of chips, so they are biased that way. FreeBSD produced excellent SMP early on compared to Linux, but unfortunately it wasn't portable enough to help the other BSDs. (OpenBSD and NetBSD are smaller projects with goals aside from application compatibility and performance, so it's natural their SMP code is developing more slowly.) Linux 2.2 kernel had some trouble going past 4, and wasn't all that good even at 4 or below, but 2.4 does much better. Pre 2.0 kernels had no SMP at all, or only severly broken patches for it, IIRC. So if you tried it again today, you should have less trouble (as long as you handle a kernel upgrade.)


    "That old saw about the early bird just goes to show that the worm should have stayed in bed."
  • BSD holds a part of the market that will always be needed. For people who need an os to run on hardware that a linux kernel hasn't even been thought about yet, you have netbsd. for out of the box security, openbsd. Linux has too many people to appease to take over these markets. It has other fish to fry rather than line by line code audits and rare hardware ports. Linux will do it's thing (haven't figured out what that is yet.=:P) and BSD will do what it does best...which is be the best in it's little field..what ever it may be, security, obscure hardware or fast web serving and file serving (if i recall freebsd is damn fast.) BSD isn't dead, it doesn't even have a cold. It's thriving and you just aren't looking.
  • They got it running with up to 6 CPUs.
    (I think I read that on DDN.)
  • Hello?

    There has been a slashdot storie about a webserver running NetBSD/dreamcast.

    http://slashdot.org/bsd/01/04/04/084213.shtml
  • Please, stop being intentionally stupid about WindowsNT,
    Please, stop being intentionally resistant to New Technology.
    it has never been, and never will be, considered a serious alternative to Unix in the enterprise level.
    That's just what the mainframe and Vax users used to say about UNIX. UNIX has been around for thirty years; NT hasn't even been around for ten. Don't write it off just yet.

    --

  • This holy war between Linux and *BSD really sucks! The main difference between Linux and BSD isn't How many processors can it use?, or My Linux box runs 12,547% faster than your *BSD box.... The main difference is Philosophy! Linux is GPL and *BSD is BSD License. Read them and then maybe will you understand what I mean.
    UNIX rocks!
  • by twitter (104583)
    Also from Berkely, the naked student. Not all novelties represent progress.

    Dude, what's that hair on my chair? Eeeewww!

  • Hehe, did they run on anything than i386 at 3.0
    at all? 8-)

    SCNR, but it seems the assumption that FreeBSD does SMP on all platform it supports is wrong. (Don't worry, I made it myself too ;-)

    - Hubert
  • Why do you say that?? It's probably because Intel talks about their typical temp and AMD talks about their max temp so AMD looks worse when in fact they are the same until you get to the P4.
  • I regularly go to NetBSD.org, and see the pathetic hodge-podge of ports they've managed to achieve. They've ported their OS to Dreamcast's and Amigas, and a whole host of obsolete boxes.

    Since when is the Dreamcast obsolete? It came out about a year ago.

    But it makes me wonder why people would expend effort banging their heads against old obsolete junk that no one is ever going to run? Old VAXStations and VMEBus junk? What masochist would even bother trying to get that stuff to run?

    What are you talking about? Have you ever stepped outside of your x86 LAN of 5 computers in your bedroom? There are a TON of VAX's still in use today, especially at colleges. I know I would rather spend $30USD to get an old VAX that does the job up and running, instead of trying to convince the university to drop thousands of dollars on new equipment. Not to mention all the people that can't afford the latest and greatest hardware (schools, new businesses, etc...)

    I wish these people would use their talents for productive things...they could be making their OS better, more stable and easier to use. Not to mention the fact that NetBSD, like the other BSD's is pretty thin on driver support for most modern hardware. Couldnt they be writing drivers for harware that matters?

    You've obviously missed the point of NetBSD in general. Their core code has to be the cleanest code I have ever seen, and that's because it has to be in order to port to all different architectures. Driver support is awesome. NetBSD even officially supports devices like the Diamond Rio 500. How about you specifically tell us what you couldn't get working.. some awful proprietary piece of microsoft hardware?

    And the whole ease-of-use thing is not something you can dismiss either...NetBSD is harder to get installed than six-year-old Slackware. I'd really -LIKE- NetBSD and OpenBSD to be more popular among users and hackers, but people like that want to program and run apps, not solve a Rubik's cube!

    First of all, you say the installation program is too 'hard', and that it turns away normal users (understandable), but you also say hackers? If something is hard, the people to pick at it first will be hackers. Second of all, have you even installed NetBSD? I think NetBSD and OpenBSD are the _easiest_ to install. It's about as straightfoward as you can get. Not to mention their new installation program makes installing NetBSD even easier.

    As for the Alpha hardware...well, Alpha has seen it's day come and go -- at least as far as hobbyist hardware is concerned.

    Alpha hardware is very alive and well. I used to work at a place doing nothing but building Alpha boxen and putting Linux on them. I can't even remember how many clusters we put together for Bloomberg. Yes, that's right, BLOOMBERG. I'm talking clusters of 100-300 nodes, all being UP2000 boards with 2 750 Mhz Alphas, and 1 gig of RAM.

    RISC along with NetBSD, and for that matter, *BSD in general, are dead.

    Dictionary.com's definition of dead (business wise): Not commercially productive; idle. So what you're telling me, is that the *BSD's have been idle, and not productive in the past few months? Ha. Far from it. Face it, youre just another Troll that uses Linux because you heard it was 'cool'. I'll put money on the fact you posted your story using Internet Explorer running some release of Windows.

    ---------------
  • I am not an SMP developer.

    but I've worked (in person) with core-level developers in both the netbsd and freebsd teams.

    I used freebsd commercially at a high-end router company who makes its bread and butter living and patching freebsd.

    and I've been lectured by those guys on how much more stable *bsd is over linux in terms of SMP and such.

    so I probably do know more about it than you, mr. hissy fit.

    I'm not sure what goal you had in mind when you flamed me, but I'd suggest you increase your medication; and you really shouldn't skip doses..

    --

  • This is a very important addition to the NetBSD kernel as the future is going to be massively paraller. But this step is not a major leap. The next thing will be addition of userland threads so that a single process may run on each processor simultaneously. And then we can rewrite all our applications to take advantage of this new feature. But why don't we all start using Linux as it is clearly superior. Technically that is. What the hell. I don't know. I think some of us like don't like GNU style of things. But thats maybe not enough. Perhaps both systems are good enough? Or maybe I want to be a kind of ultra weird nerd for who Linux is too much of a common thing. What else. I hope my NetBSD box mutates into a flying saucer. As I think I need to take a trip to an other planet quite soon. KICK ASS
  • You mean like the assumption that Linux does SMP on all platforms like the post I was responding to assumed? :)

    And I still need to check if Linux does SMP on Alphas anyway - I seem to recall it does, but I'm not 100% sure.

  • Why is it that the people that post the most obnoxious, ignorant posts all do so anonymously?

    Coincidence?

    If NetBSD is dead, why can they do things no other OS can do? Show me a Linux distro that runs flawlessly on two different processor architectures. I'll show you a more stable, more robust, and more secure OS that runs on 21!

    Putting down BSD doesn't change anything - people will still develop it, ISPs, companies, and people will continue using it, and we'll all be happy.

    Go see http://www.sixgirls.org/ and see what an obsolete Amiga can do with NetBSD. 100 users, more than 50 domains... What're you running on your Windows box?

  • No, not quite true. FreeBSD [freebsd.org] supports multiprocessor x86 and has for awhile, NetBSD [netbsd.org] is rolling out SMP on x86, alpha, sparc, and ppc (see this post [slashdot.org] for links.

    NetBSD took longer, because portable code that will support many platforms easily is their grail. Writing that type of code naturally takes more time.


    "That old saw about the early bird just goes to show that the worm should have stayed in bed."
  • It's not a matter of a fixed number of spinlocks. It's a matter of the whole kernel being *designed* to be run on more than one CPU, and only locking those resources that are absolutely needed to be locked and only for the minimum amount of time for which they need to be locked. One Big Kernel Lock fits neither condition. It's a kludge, designed to let marketroids (and yes, free OSes have them, too, even if they don't get paid for it) claim "SMP support" when in fact it's a poor imitation at best.
    --
  • by ernop (111430)
    The Linux "fine-grained" locking is embarrassingly often just a One Big Kernel Lock. Read the source, you'll notice...
  • lol, wow I guess you are right -- now where did I put that medication?

    But seriously folks, you are still just talking out of your bum. Please I don't agree or disagree with you but some content would be wonderful. Word of mouth doesn't mean crap to me. So you have heard some people say somethings. Doesn't mean much. Why don't you find out some facts and post them. You say that 5.0 will bring FreeBSD equal to 2.2 Linux in SMP support. How, what have they added -- what have they ignored that makes it lag behind 2.4? Just what are they doing wrong that Linux is doing right? And are their goals even the same?

    I am willing to admit that Linux probably has better SMP support than FBSD. I would admit this because FBSD is much more likely to provide a stable solution that is a little behind than a cutting edge solution that is not stable. Personally, I agree with FBSDs priorities myself -- but that doesn't make it better. Show me something as proof of your claim.

    I am proud that you probably do know more than me. But you could show it by posting some of that knowledge and not just making unsupported claims. And my goal when I "flamed" you, was to try and encourage you to do just that. Support your statements. If you are going to pretend you have a good reason to feel a certain way, share that reason and be prepared to back it up.

    ---
    "Do not meddle in the affairs of sysadmins,

  • Usually NetBSD starts porting first, then other OS'
    (including *BSD and Linux) developers take over
    its code and set a bunch of developers onto it
    in order to complete the progress.
    In case of *BSD this doesn't matter.
    In case of Linux, they have the f... GPL, so
    NetBSD people (and *BSD in general) can NOT
    profit from the Linux peoples' efforts, but THEY
    "stole" NetBSDs code (perfectly legally).

    IIRC Linux developers have begun to dual-license
    their code: GPL for the kernel, BSD for *BSD.


    --

news: gotcha

Working...