SMP On OpenBSD, Coming Soon

Lord of the OpenBSD writes "At long last, SMP development on OpenBSD looks to be gearing up. One person is now doing full-time funded development on SMP. Project leader Theo de Raadt is now asking for funding for a second developer. Theo has announced that SMP support for i386 is planned for the OpenBSD 3.6 or 3.7 release, the first of which is due in 8 months."

Yet another modern feature added to *BSD

[Anonymous Coward]

Yet another modern OS feature is being added to *BSD, which have many features not even found in the best of commercial operating systems. *BSD isn't dying, it's setting the standard for other operating systems to follow.

Re:smp?

[Anonymous Coward]

SMP is one of those technologies encumbered by various patents. As it is now, SCO claim to own most of the support for it, and it's one of the features embroiled in the SCO vs IBM case.

How they'll get around this, I don't know. It's good to see the coding and experience getting out there and used all the more however.

Re:smp?

[TheHonestTruth]

Because it wasn't important to Theo. Seriously. He had no need for it, plus it introduces security issues (I guess, I can't speak from experience) with what code is getting executed in what processor, so it wasn't developed for a long time (security being OpenBSD's focus). It just started getting some work in the past year or so.

-Truth

Re:Interesting...

[Ryvar]

While I'm a big fan of OpenBSD and Theo both - I have to admit that Theo doesn't play nice even with those NOT beneath him - the loss of Niels Provos is still a bitter, bitter blow for the project.

irrelevent

[quelrods]

I fail to see how this is relevent. This whole article is just going to be eaten up by the /. trolls. SMP support has been coming along in openbsd for the last year+, mostly "borrowing" code from netbsd. Yes SMP is coming and it will bring them up to date with the rest of the world. There really isn't anything to see here...move along.

it's not "porting"

[Anonymous Coward]

This is not just "porting" like a device driver.

SMP touches every aspect of the kernel (scheduling, VM, VFS, etc.). Each OS is different internally and so you can't just rip code out of one and put it into another. It's not simply copying over a sub-directory and changing a couple of kernel system calls.

You have to pour over a lot of the files and make all the data structures are written to and read from correctly.

There's also more than one way to do SMP so how do you know whether he's "reinventing the wheel", or coming up with a novel approach?

Re:SMP is good, but what about pkg management?

[Anonymous Coward]

OBSD has both ports and packages (precompiled binaries) and the software to manage them, pkg_*. When it comes time to upgrade your OS, all your customizations can be put in a site.tgz file. Takes one command to install it.

There's a reason OpenBSD has nice man pages and FAQ - they're for learning how the OS works.

Re:smp?

[Daniel_Staal]

OpenBSD's top priority is security. For SMP that means two things:

1. All potential security-relevant race conditions must be handled. A single processor system can never do two things at exactly the same time. A dual processor one can. OpenBSD wouldn't be OpenBSD if that would be allowed to affect the system's integrity.

2. Given the choice of an small project, that increases security, and a big one that probably will lower it, Theo will choose the one that increases security. Dual-processors are not a major concern to OpenBSD's core users, so support can wait until other things get done.

Re:Interesting...

[Daniel_Staal]

Let's see...

Privsep: Makes sure that only the code that needs to be root is run as root.

WorX: Makes sure that data cannot be executed as code. (Memory pages are either executable or writable, not both)

There are other features like them, that add to the security of the system. OpenBSD is better at firewalling than FreeBSD because of them; it is harder to break any port or access that is allowed/found into a security lapse. And any security lapse will be as limited as possible.

FreeBSD is descent at this. The standard level of security is available. OpenBSD is positively paranoid. That's a good thing when anyone can throw anything at it...

Felix von Leitner's "Benchmarking BSD and Linux"

[mosel-saar-ruwer]

I wish I could refute this, but IIRC, a series of I/O benchmarks were run on the major OS players a while ago and OBSD did pretty terribly.

The money quote from the Conclusion of Felix von Leitner's Benchmarking BSD and Linux:

OpenBSD 3.4 was a real stinker in these tests. The installation routine sucks, the disk performance sucks, the kernel was unstable, and in the network scalability department it was even outperformed by it's father, NetBSD. OpenBSD also gets points deducted for the sabotage they did to their IPv6 stack. If you are using OpenBSD, you should move away now.

http://bulk.fefe.de/scalability/ [bulk.fefe.de]

http://developers.slashdot.org/developers/03/10/19 /0130256.shtml [slashdot.org]

Re:Yet another modern feature added to *BSD

[Santana]

As you may know, OpenBSD focuses on security. SMP support brings new concerns on this field.

Re:Is this necessary?

[Noryungi]

What's really lagging in OpenBSD is an easy to use port/package system

I am really sorry but have you even used OpenBSD recently? I installed OpenBSD 3.4 last month on a small server at home and installing third-party software was as simple as:

For a package:

cd <path to packages>
pkg_add <name of package here>

For a port:

cd <path to port>
make install

And... that's it!

Could you please explain to me how this is difficult?

Re:Yet another modern feature added to *BSD

[anthonyrcalgary]

OpenBSD is hardly the leader of the pack as far as performance goes. Even on UP systems, it's still slower than almost everything else in key areas (disk performance being the big one). When it has SMP support, it will initially use one big mutex to lock the kernel, and will not initially be optimized for anything weird (Hyperthreading, NUMA). Sure it's my favorite OS. But not for everything.

Re:What ever happened to..? Water..who?

[Thelonious Monk]

Here is the OSNews link... http://www.osnews.com/story.php?news_id=2320 I knew i wasn't on crack.... today.

Re:What Will Theo Use Processor 2 For?

[Imperator]

Well SMP stands for "symmetric multi processing". That basically means the kernel can run on any processor--they're symmetric in that respect. (The advantage of being symmetric is that multiple processors can run kernel code at once, but the disadvantage is that you need locks and the like.)

If he were going to use the extra processors for nothing but crypto, (a) he'd be wasting them since crypto doesn't take that much CPU by today's standards and (b) it wouldn't be called SMP.

Re:Yet another modern feature added to *BSD

[dmiller]

SMP isn't just a bit of software that you can port from one OS to another. It touches just about every kernel internal and changes many assumptions on the way. That being said, the approach to implement SMP in OpenBSD (and some code) is being derived from NetBSD.

Re:Yet another modern feature added to *BSD

[bccomm]

Mod parent up, he made some good points.

However, I'd have to disagree with FreeBSD being the technological leader of the bunch. It's an excellent system, and is the most widely used/commercially supported of the three (or six, with ekko, DFly, and Darwin). However, I see NetBSD being much more advanced for a few reasons:

1. It's more clean. ``It isn't done unless it's done right.'' As a result, it's much easier to extend (new drivers, new archetectures, ...)
2. It is the first free unix-ish system to have many new features, like USB, IPV6, and crossbuilding support (ROCK is the only other one I can think of that has this) just to name a few.
3. As a result of #1, it can serve as an excellent resource (a reference platform or nice collection of example code to stare at when you're bored).
4. It's small, but generally highly scalable
5. ...

I thought the very same way as most users at one time. I used to be a devout FreeBSD user. After buying a bit of HD space (bit=320GB), I decided to take on the multiboot challenge. I installed a total of thirteen different unixes (no windoze), telling myself to create the *exact* same environment on each. I decided to give NetBSD an OK-sized 10GB partition. The next day I swapped the 60G I gave to Slackware with it. It was faster and seemed, generally, a whole lot cooler. Within a week, I had a nice, stable NetBSD-current system up and running and found myself not being able to reboot to finish installing Solaris, OpenBSD, and Gentoo!

The point is this: NetBSD is the `forgotten' unix in many ways, and I, for one, find that sad. I think all the BSDs, along with Linux, will be around for some time. NetBSD, though, is simply the bliss that I, too, nearly overlooked.

Re:Yet another modern feature added to *BSD

[Just Some Guy]

Add to that the fact that a BSD system will not automatically upgrade your /etc, then you have the best reasons that say, a Debian box is easier to maintain.

Qualify that: FreeBSD has had "mergemaster", which semi-automatically upgrades your /etc (and even walks you through merging changes) for a long time. I'm not sure with Open and Net haven't imported it yet, or at least hadn't when I last looked, but at least one BSD currently enjoys easy /etc upgrades.

Re:Yet another modern feature added to *BSD

[grub]

All that crypto and random goodness doesn't come for free.

True enough, but with a ~US$90[0] Soekris crypto accelerator [soekris.com] it's damn close :) Nothing like having a heavily used IPSec tunnel with your CPU being relatively idle.

[0]- when I bought mine in mid 2002

Re:SMP is good, but what about pkg management?

[Anonymous Coward]

> I remember having to manually edit files in
> /etc, and having to figure out which files were
> added or deleted since the last version.

try mergemaster. when you become better accustomed to the process and have decent tools, i think you'll prefer it to debian.

cd /usr/ports/sysutils/mergemaster
sudo make install
sudo mergemaster

Re:What Will Theo Use Processor 2 For?

[evilviper]

but I thought his other idea was cooler. On-the-fly encoding and decoding and hiding of jpegs from wives and whatnot.

I really can't see the point. For $100 you can buy a PCI crypto card that would do 3DES as fast as most would ever need.

Or they can just act as an incredibly fast random-number generator (something CPUs aren't very good at) if you are doing some crypto that the card doesn't support (blowfish isn't popular in hardware, yet).

Re:roots

[Tet]

Linux doesn't run on as many platforms as NetBSD

Myth. Linux does (and has for many years) run on just as many platforms as NetBSD. Most of NetBSDs "platforms" are actually just variants on a single architecture. Thus while NetBSD counts atari and amiga as separate ports, Linux just treats them as part of a single Linux/m68k port. In fact, NetBSD runs on two architectures that aren't currently supported by Linux (ns32k and vax), whereas Linux run on five that aren't supported by NetBSD (mips64, ppc64, s390, sh4 and etrax). I'm not trying to put down the worthy efforts of the NetBSD community, but I just get a bit fed up with people claiming that it's more widely ported than Linux. It was true in the past, but hasn't been for some time.

Re:'asking for funding'

[don_carnage]

Err...wasn't that a DARPA grant?

Re:roots

[sir_cello]

It is probably true that Linux does run on more systems than NetBSD, but the support is fragmented and disparate at the best. This is the essential and important distinction.

NetBSD ensures that the one overall "package" (kernel + user space) works equivalently across a set of platforms. Your installation (executables, directories, config, etc) are largely equivalent across all platforms: take your custom scripts and system setup and find that it can be dropped onto NetBSD/other with little cost.

This is definitely not the case with Linux as each platform largely a different and somewhat incompatible distribution.

Linux strength is that you can find a large and interesting variety of distributions for all sorts of specific niches and purposes. It's weakness is that you can't find the one distribution that works across many platforms. And this is the niche that NetBSD has.