Submitted by “Matt Dillon, one of the FreeBSD kernel developers, decided that several of the approaches being used in the 5.x series were dead-ends, and in July 2003 forked the stable 4.x codebase to form DragonFly BSD. The 4.x FreeBSD Foundation meant that DragonFly has been a solid platform from the start. DragonFly, like the other BSDs, imports code from other members of the family when it makes sense, such as the malloc() security features from OpenBSD, parts of the WiFi subsystem from FreeBSD, and USB code from NetBSD. In spite of this, development has been pushed in some unique directions.”
Do we need to hear why Matt Dillon left FreeBSD everytime someone wants to publish something related to Dragonfly?
Maybe because if people don’t hear it, they spread FUD and this is very bad for such a small but healthy community like in DragonFlyBSD.
Or may be such statements are FUD. like “dead-end”?
I fail to see how FreeBSD’s approach became a dead end! Everyday I hear good news about their excellent new features and better cpu scaling and performance (GIANT-Free code, new network stack, new scheduler, SMPng, etc) look at recent mysql results…
I think it should be viewed as a comparitive approached (so all can benefit) instead of insulting it by calling it dead.
I do not have anything against Drangfly in fact I love it. It just caught my eye when someone wants to praise DrangonFly they usually denounce FreeBSD.
Nor do I have anything against Matt Dillon I still see him posting in FreeBSD mailing lists which just shows that FreeBSD is not a dead-end otherwise he wouldn’t share be posting there.
Thanks for their great work after all.
Matt Dillon…decided that several of the approaches being used in the 5.x series were dead-ends
Someone’s opinion is not FUD.
The main reason we call it ‘computer science’ is that we are still experimenting. So the more ideas on different approaches is very important.
Great, so you take one sound bite and take it completely out of context – damn i hate those who come in at the end of the conversation and trying to add their 5cents worth to the discussion.
Matt Dilon never said it was going to be a ‘dead end’ as in it would be impossible; his argument is that he can achieve the same sorts of results as FreeBSD 5.x (and subsequent releases) using a different method to going about achieving it.
His arguement was that SMPng is possible, but it is more complicated that it needed to be, and doesn’t address areas which he wanted addressing – so he split of to take development in a whole new direction – IIRC it was an IBM’er here who actually pointed out that their approach to scalability is very similar to a project done by IBM over a decade ago.
I didn’t take it out of context. I just commented on that part and I was specific I believe… you do not have hate
it’s not good for your health. My argument was clear I believe you misunderstood what I was trying to say.
I am aware of what Matt said! Thanks for adding that.
Do we need to hear why Matt Dillon left FreeBSD everytime someone wants to publish something related to Dragonfly?
No, but the media thinks we’re stupid, so they translate more complicated situations into black and white, us versus them. They’ve been proven right over and over again, so why should they pander to the intellectually sophisticated minority? We’ve already made up our minds, and there’s nothing they can do to convince us that they’re not complicit in the New World Order. But I digress…
It’s a fallacy to believe that just because an operating system is free software, it has to be a universal operating system for all systems and all purposes. Freedom implies the opportunity to challenge the prevailing notions and in the process, contribute exciting new ideas. We can benefit from each other’s work without agreeing every step of the way, and DragonFly is a great example.
Just about everybody that speaks ill of DF fails to understand the goal of the project. Most UNIX-like operating systems are designed to support UP, SMP, and possibly NUMA systems. DF is designed to support clusters of such systems. It’s designed under the premise that maintaining cache coherency across nodes in a cluster can be prohibitively expensive unless this issue is directly addressed in the OS architecture.
Personally, I feel that DF kind of misses the point. We have virtualization. Why do we need to port applications to a new OS in order to run them efficiently on a cluster? I think the goal should be a clustering hypervisor that exports SMP virtual machines on which most modern operating systems can run side-by-side. In other words, a mainframe OS that runs on a cluster of inexpensive computers. DF could become such a system, and in fact, it would be easier than dealing with the hassle of directly supporting existing applications.
If DF is what the IT industry wants, then Plan9 would be alive and kicking today, and DF wouldn’t exist. They don’t want a new UNIX for clusters. They want a mainframe for clusters, on which various industry-standard UNIX systems (and/or Windows) could be deployed. But, tragically, IT can’t articulate what they want unless it is spoon-fed to them in a glossy marketing brochure.
You made some interesting points here. I may reply:
“It’s a fallacy to believe that just because an operating system is free software, it has to be a universal operating system for all systems and all purposes.”
I’d go a bit further. Users – in which class IT managers can be seen, too, because they do not know what they want or what they need – do not use the OS, they use applications that run on the OS. But for them, OS and apps are all the same. And if it’s free altogether, it’s supposed (!) to be fitting all present and future expectations. In Germany we call this “eierlegende Wollmilchsau” (egg laying wool milk sow), you surely know the term “one size fits all”. But as we know from reality, this usually does not work in the computing world.
“Freedom implies the opportunity to challenge the prevailing notions and in the process, contribute exciting new ideas. We can benefit from each other’s work without agreeing every step of the way, and DragonFly is a great example.”
I completely agree here. They have great ideas which, maybe, seem to be obsolete today (for example, due to virtualization), but the approach they’re taking is interesting, at least from a scientific or educational point of view. I won’t say DFBSD is a “one size fits all” OS, but it will have its place in applied IT, I’m sure.
“DF is designed to support clusters of [ UP, SMP, and possibly NUMA ] systems. It’s designed under the premise that maintaining cache coherency across nodes in a cluster can be prohibitively expensive unless this issue is directly addressed in the OS architecture.”
Maybe this (a bit technical) point of view is too “geeky” for some and does not fit into today’s buzz word corrupted media? 🙂
“Personally, I feel that DF kind of misses the point. We have virtualization. Why do we need to port applications to a new OS in order to run them efficiently on a cluster? I think the goal should be a clustering hypervisor that exports SMP virtual machines on which most modern operating systems can run side-by-side. In other words, a mainframe OS that runs on a cluster of inexpensive computers. DF could become such a system, and in fact, it would be easier than dealing with the hassle of directly supporting existing applications.”
Hmmm… this is a valid point. Virtualization “saves time” on the application developer’s side, but consumes more time on the OS developer’s side, and it usually comes along with a certain overhead which may eat up performance advantages. (NB: You cannot “save time” anywhere.)
We had this in 70s on the IBM big irons… 🙂
“If DF is what the IT industry wants, then Plan9 would be alive and kicking today, and DF wouldn’t exist. They don’t want a new UNIX for clusters. They want a mainframe for clusters, on which various industry-standard UNIX systems (and/or Windows) could be deployed. But, tragically, IT can’t articulate what they want unless it is spoon-fed to them in a glossy marketing brochure.”
Nothing to add here, completey right. (See my statement about IT managers above.)
“It’s a fallacy to believe that just because an operating system is free software, it has to be a universal operating system for all systems and all purposes. Freedom implies the opportunity to challenge the prevailing notions and in the process, contribute exciting new ideas. We can benefit from each other’s work without agreeing every step of the way, and DragonFly is a great example.”
Good points and in general I agree with you.
“Just about everybody that speaks ill of DF fails to understand the goal of the project. Most UNIX-like operating systems are designed to support UP, SMP, and possibly NUMA systems. DF is designed to support clusters of such systems. It’s designed under the premise that maintaining cache coherency across nodes in a cluster can be prohibitively expensive unless this issue is directly addressed in the OS architecture.”
I also largely agree with you here.
“Personally, I feel that DF kind of misses the point. We have virtualization. Why do we need to port applications to a new OS in order to run them efficiently on a cluster?”
I am not however sure that I agree with you here. Although DragonFly’s developers are doing much rearchitecting of the kernel core to natively support clustering, fact remains that it *is* in fact a general purpose OS, providing many third party apps (through NetBSDs PKGSRC), and provides a number of drivers to support hardware functionality that in many (if not most cases) would seem largely out of place in a system designed purely for clustering.
Also I am not convinced that the effort needed to port applications to DragonFly to efficiently support a largely distributed system would be any more involved than porting such apps to any other distributed system, and in some cases (due to the fact that much of the new functionality one will see from DF once the MP lock is gone, and the syslink stuff is finalized) is largely transparent to things in userland. Also when one takes into account binary compatability with Linux and FreeBSD, that would seem to lessen the headaches involved with porting even more, perhaps to the point of eliminating it in some cases.
“I think the goal should be a clustering hypervisor that exports SMP virtual machines on which most modern operating systems can run side-by-side.”
Perhaps this could be useful, but I am not sure if this would end up being more than less effort than what is going into systems like DragonFly.
“If DF is what the IT industry wants, then Plan9 would be alive and kicking today, and DF wouldn’t exist. They don’t want a new UNIX for clusters.”
I am also going to have to disagree with you here. There are certainly a number of places where clustering of various sorts is desired, and in many of those cases by people who are familiar with Unix-like systems, or more specifically have some manner of investment in Unix-like systems. Having a such a system that in addition to being useful as a general purpose OS which also has *native support* for clustering could well be a time and or money saver for those people.
“They want a mainframe for clusters, on which various industry-standard UNIX systems (and/or Windows) could be deployed.”
Although I am sure that there are indeed a number of people/businesses that want or require such, I am equally sure that for others something like a future version of DragonFly running on commodity hardware would be an equally fine solution.
Final notes,
I am indeed a fan of the DragonFly project (despite my ocasional grumbling regarding the rate of progress in some non-core areas), and I would definately like to see it succede in the years to come.
With only a small group of developers, they’ve done many things that have taken other projects either longer or more people to do, and have (IMO) done so in a more generally thoughtful way.
I wouldn’t bet money on it taking the world by storm anytime in the near future, but in addition to being an interesting OSS project, I can see it having a future no less bright than that of the other BSDs.
Claiming FreeBSD options, in the multi-threaded department, are a dead end does not mean, necessarily, that it’s not a valid project !
It doesn’t mean a negative criticism either. Both communities should interchange opinions and visions. Hence, it’s normal to see, friendly, developers posting in the same mailing lists ! This is NOT Linux ! (Sorry for plenty of exclamations signs).
Remember that BSDi gave some training on SMP (Symmetric Multi-Processing) to the FreeBSD team where there were a lot of divergent opinions, right away, and it became evident along the years…
In my humble opinion, FreBSD has not fullfilled the needed multi core (CPU kernel) tasks that the future brings. It’s also a “not so good” implementation for the multi-threaded future; as it just «imitates» the common approaches to multi-threaded kernel when it could be more “fearless” in design… (design comes before implementation).
DragonFly, at least, is trying a new approach.
Check this
http://www.zdnetasia.com/news/hardware/0,39042972,62021390,00.htm
The necessity of a new (clustering) file system was detected and, immediately and fearlessly, announced. That’s development…
That’s what I would call the “crystal clear” approach to problems from free open codding.
In my humble opinion, FreBSD has not fullfilled the needed multi core (CPU kernel) tasks that the future brings.
Well, recent performance numbers where is scales better than Linux for multi-cpu systems shows that be an incorrect conclusion. SMPng for FreeBSD was a long project but the fruits are beginning to show very nice promise.
This does in no way invalidate what Dragonfly is doing as the goal is simply very different
DragonFly looks extremely promising. VMS-style clustering, coupled with jails would make for an extremely pervasive hosting environment. Even more so if you throw Linux binary compatibility into the mix.
Perhaps DragonFly/VMS-style clustering will form the basis of the next generation of hypervisors: pooling resources within a cluster and dishing them out to the VMs (or pseudo-VMs) that run on top of it.
NetBSD is irrelevant. FreeBSD is maintained by incompetent people (according to linus, at least 🙂 and OpenBSD has maniac leadership.
So I welcome DFLY
NetBSD is not irrelevant at all!
On old hardware and on small computers it is sometimes the only true option you have.
NetBSD’s pkgsrc package management infrastructure is the DragonFly’s default package management software; BeleniX also uses it as its default package manager subsystem.
Edited 2007-06-18 22:09
Well, this co-founder of NetBSD seems to think it’s irrelevant:
http://mail-index.netbsd.org/netbsd-users/2006/08/30/0016.html
@predictor, Well if Linus said so, It must be true…:P
Really though, I’m using OpenBSD right now.. but I have great respect for all of the BSD projects, And there has always been great deal of sharing going on between them.
Linus made a recent claim about lack of “merging” in the BSD community, What an idiot.
They are all unique projects, Each with different goals and ideas.. but they still continue sharing code where it makes sense.
Edited 2007-06-18 18:41