Matthew Dillon discusses software innovation in this short post on the DragonFly kernel List relating to the forthcoming journaling layer. Elsewhere, I/O Consolidation and Direct-to-DMA plans.
Matt Dillon on innovation and DragonFly
About The Author
Eugenia Loli
Ex-programmer, ex-editor in chief at OSAlert.com, now a visual artist/filmmaker.
Follow me on Twitter @EugeniaLoli
Hi, link to the dragonfly site is broken, it links to:
osnews.com/www.dragonflybsd.org
For further thoughts on researc/innovation in OS’s, read
http://www.cs.bell-labs.com/who/rob/utah2000.pdf
I love DragonFly and like to see anything relating to it’s current innovation. This article is really interesting!
Go read the mailinglist archives. DragonFly BSD is a major redesign. It’s UNIX done right and it will be huge. Yes, it’s still BSD, which means you can actually use it today and get things done.
There is a stable cvs tag (and ISO images to get you started) which lets you stay close to progress without being at the absolute bleeding edge.
The ISO itself is very small. The installer is better than the old FreeBSD installer, but it’s still text only. It’s a full BSD with good documentation. Don’t expect to be pampered though. If you expect a Mac/Windows experience it’ll be like being dropped on the moon, albeit with a great toolbox and a huge manual. You’ll figure it out.
jonas.kirilla posted….
It’s UNIX done right
I am curious, what does this comment mean.
I still don’t see how the ‘3 legs’ he describes are directly ‘visible’ to the end-user.
What does KVA stand for?
Also what is an SFBUFs or MSFBUFs?
Looks like an ambitious project to push BSD and Un*x forward. The developers seem capable… I wish them luck & courage.
“Not once did I ever get any other answer. Pretty sad, I think, and also sadly not unique to FreeBSD. It’s a problem with mindset, and mindset is a problem with our educational system (the entire world’s).”
A couple of observations.
1) Why is he mentioning FreeBSD. If you have valid points, no need to take someone’s name and be negitive. Its fine when you say FLOSS or Commercial but to ID some group is somewhat wrong.
2) If FreeBSD is so broken, then why use their source code. Be orginal and create your own.
3) Heck why even do Unix, its over 30 years old in design.
This is a fork of FreeBSD, and he explains the reason. He had a different vision of FreeBSD, but he couldn’t convince the other developers so he had to go his own way. Some bitterness is understable I think, it doesn’t mean FreeBSD is broken.
I say it’s “Unix done right” (my term) since I believe DragonFly will be better than the other BSDs, and better than Linux. (Big Iron is out of my reach, as are the Unixes that run on such hardware.)
It’s not just this or that feature, but all of them. Look at the goals and what has already been accomplished.
“2) If FreeBSD is so broken, then why use their source code. Be orginal and create your own.”
Your argument is flawed. Matt objected to some of the design decisions in FreeBSD 5.x. DragonFly BSD is a fork of the previous 4.x.
Matt was a FreeBSD developer. He knows the source very well. Part of it *is* his source.
3) Heck why even do Unix, its over 30 years old in design.
Consider this: why does every “hobby” OS project out there use ‘gcc’, and a gazillion of GNU utilities? And what kind of runtime environment do these utilities expect? Something POSIX-ish at least. Sure, it’s possible to create non-Unix platforms from scratch, but you’re in for a world of hurt, or at least a very long and rough ride. Don’t expect to have KDE or Gnome up and running any time soon.
BTW, cars are old too, as is eating meat, or having sex.
If someone actually read the redesign it is basically implementing features already in Linux, or features deemed unnessicary, into a BSD liscensed product.
DragonFlyBSD’s design goals have much in common with AmigaOS meets SSI clustering than anything the proprietary GNU/Linux systems have to offer. Some of its features are unique and no other OS provides them.
I have used FreeBSD for years and I am very interested in the progress DragonFly is making, although its not clear why people would get off the linux bandwagon at this point. What are the clear and obvious differences for users?? (even the highly technical userbase that might use DragonFly) I don’t see a very compelling reason to move given the architectural changes Matt is implementing.
People use FreeBSD because of stability. DragonFly will have to mature greatly before users have this perception (note I said perception, which is not to be confused with fact).
People use linux so they can be on the right side of the trend…vendor support, ability to get precompiled packages for any major codebase etc. DragonFly clearly cannot match this.
I see DragonFly as being the server BeOS. A brilliant system that truly meets its design goals but goes nowhere due to bad timing. The market offers an opportunity for a new architecture to take root perhaps once a decade. The window (pun inteneded!) opened and linux was there to take up the mindshare. I don’t see the linux bandwagon getting derailed anytime in the next four years. At that point another technology will have a chance, but not now, people are too busy with linux.
Here are four very good links to read:
http://www.shiningsilence.com/dbsdlog/archives/000331.html
http://www.shiningsilence.com/dbsdlog/archives/000428.html
http://www.shiningsilence.com/dbsdlog/archives/000519.html
http://en.wikipedia.org/wiki/DragonFly_BSD#Additional_features
…to design the kernel for the multi-core future. At least that’s what I heard.
1) Why is he mentioning FreeBSD. If you have valid points, no need to take someone’s name and be negitive. Its fine when you say FLOSS or Commercial but to ID some group is somewhat wrong.
Because it’s based on FreeBSD 4.x because he didn’t liked where 5.x where going?
2) If FreeBSD is so broken, then why use their source code. Be orginal and create your own.
Because the FreeBSD 4.x code was a good base?
3) Heck why even do Unix, its over 30 years old in design.
Because starting over from start whould take more time?
“BTW, cars are old too, as is eating meat, or having sex.”
And two of them is wrong ;D
Before the FreeBSD 5 branch went stable, the DragonFly team would say that FreeBSD won’t ever go stable because the solution the FreeBSD team took was too messy and too difficult with very little payback. The also said that the kernel has become an unmantainable mess. I don’t hear that anymore. I’m not sure if it is because FreeBSD has gone over the major hurdle and cleaned up considerably or if it is for another reason. Anyways, has anyone looked into the FreeBSD kernel and done any studies on how clean the code is, how difficult it is to mantain, how difficult it is to extend, etc? I myself, in the userland, enjoy FreeBSD immensely and have found 5.3 extremely stable and capable. But I am clueless in regards to studying the kernel structure and code, so I would love to see some kind of unbiased third party analyzation of the FreeBSD kernel and see where it is now, what is wrong with it, and where it should be headed for.
“Consider this: why does every “hobby” OS project out there use ‘gcc’, and a gazillion of GNU utilities?”
Because people don’t have any new ideas? Because the only type of OS they were taught about at University was Unix?
There is a real danger of coming out at the end of a course thinking what you had to learn was all true.
Matt Dillon?
http://www.imdb.com/name/nm0000369/
This Matt Dillon ?
“Consider this: why does every “hobby” OS project out there use ‘gcc’, and a gazillion of GNU utilities?”
Because people don’t have any new ideas? Because the only type of OS they were taught about at University was Unix?
There is a real danger of coming out at the end of a course thinking what you had to learn was all true.
From my experience, most colleges teach using Visual Studio or Borland(a few years ago) and focus a lot on Windows. Some of the higher level courses might include unix, but that depends on the university. This is especially true of community and junior colleges. At the bigger and more prestigious schools, you’ll probably see a lot more Unix type stuff, but it’s not the focus.
I don’t think there is an indoctrination of “gcc is the best” at the university level. Gcc is used because it is free (open source and cost wise) and can target multiple platforms. Really, it’s practical concerns and has little to do with indoctrination, although I do agree that there is a danger of believing everything that is taught to you (ie my professor told me that gcc was a terrible compiler).
@Ecio: This Matt Dillon:
http://apollo.backplane.com/
> People use FreeBSD because of stability.
This is one of the major reasons why I use DragonFly.
FreeBSD 5.3R is just too unstable and does not meet the
goal of a turning off the Big Giant Lock, so why pay
for all that performance loss relative to FreeBSD 4
and to the FreeBSD 4-based DragonFlyBSD? DragonFly
is just as fast as FreeBSD 4 and it has a clear path
to MP scalability, whereas FreeBSD 5 seems fubar’ed
beyond repair.
> Before the FreeBSD 5 branch went stable
Just because they call it stable does not mean it is.
The freebsd mailing lists
are full of crash reports and instability reports.
FreeBSD people in the know like John Baldwin has stood
up in conferences at BSDCon and stated there’s no way
his company would use FreeBSD 5 instead of FreeBSD 4
in a production system.
. . . sounds very interesting. I’ll be watching this with interest.
Because people don’t have any new ideas? Because the only type of OS they were taught about at University was Unix?
Because a lot of the ideas in Unix simply make sense?
I think DragonflyBSD’s installer is a great new thing for the BSD community and I figure it will get ported to other OS’s like FreeBSD.
I don’t understand why it needs to be graphical though… I personaly like text but I work with servers so what do i know. i would figure there would be a higher install base for servers than desktops.. like linux and freebsd.
Someone posted a link to a Bell Labs paper dealing with the lack of innovation in the software industry. It’s mostly a blurb against widely used operating systems, while hinting that Plan 9 is the real deal.
Well, I’ve checked that OS and it doesn’t look great : the installation is done through the command line (as spartan as OpenBSD); the desktop looks worse than DOS 3.11; the hardware support is minimal, etc. Four years after that paper was published, Plan 9 is still a toy for graduate students, not something I would use as a server and certainly not as a workstation
I don’t mind text installers. I’m by all means no expert. I’ve installed OpenBSD a few times, and it’s pretty straightforward, but you do need to RTM first, and when you’re done, you’re pretty comfortable with some tasks which appear to be daunting if you’ve never done them before. I guess ppl who prefer to “wing it” through installations balk at that, but it’s really not that bad. The instructions are very clear and well-written.
I can only say that FreeBSD 5.3 is not stable as expected. The truth is that there’re some key problems to be solved, but it’s clearly that 5.x is going to get the stability. Many new features were implemented in to the 5.x and 6.x trees. If you don’t need the new stuffs, stay with 4.x (4.11-RC for now).
I really don’t like the f@cking Matt’s behaviour. Please read back the freebsd-hackers mailling-list then, you’ll catch out why this Matt lost his commit bit.
So he started his new project, DragonFly. It’s so kind. We have new member in the BSD family. But the way Matt and some young DFly developers talk about FreeBSD project is simply can not be accepted, there’s no conscience at all.
>> This is one of the major reasons why I use DragonFly.
>> FreeBSD 5.3R is just too unstable
more to stability than the kernel
the build system
the ports tree
etc etc
maybe dragonfly has surpassed freebsd in these respects, but it is unlikely freebsd shops will be quick to switch. heck, the fact that they haven’t moved to linux shows how stuck they are on freebsd stability.
Thanks for the links bsdrocks.
＞ more to stability than the kernel
＞ the build system
＞ the ports tree
＞ etc etc
DragonFly runs FreeBSD binaries! It uses the FreeBSD ports system! As for the build system, just type “make” in /usr/src. Simple!
It was open sourced so users can make it something better and bigger.
I’ve installed and run it on one of my boxes for a little while. Not bad. Looking forward to the point in time when they begin focusing on packaging and stuff like that so that I will be able to have my cake and eat it too.
I found his blurb on moving the journaling into a kernel layer/module rather interesting. Not entirely sure I understood the implications. For example, what does that mean for existing, journaled file systems? :/ (Maybe I should reread it tomorrow when I am not so tired).
I am hoping all the innovative (for BSD) rewrites he is making will add up to a spectacular performer that outperforms existing BSDs and Linux.
Until the “whole package” is ready tho’ I will still to my current 3: Mac OS X, Ubuntu and BeOS.
Mike
Regarding the how-does-the-kernel-source-code-look I actually have no idea, but if you or anyone else whould want to take a look and see how things are done I suppose NetBSD would be a much better start since it’s kind of their goal to keep the code clean and portable.
I also installed it earlier and actually still have it installed althought I doesn’t use it longer. The reason was my ports system broke. And now I’m not sure which way to go, I’m considering:
Yoper Linux: Highly optimized, fast, no-work (precompiled binaries) Linux distribution. I would have the fastest system, good nvidia drivers, could run winex/cedega, and so on.
FreeBSD: Tried classic, using ports is kind of a waste of time thought, and portupgrade never works as good as I would want it to work. I know there is Gentoo aswell but you “have” to rebuild everything for each and every of their useless patches so no thanks, I would still have nvidia drivers.
DragonFly: I would prefer using it over FreeBSD, I would still have the nvidia drivers and so on but I’m not confident my system would never broke due to the large changes. There is the GoBSD distribution but well, I’m not confident.
NetBSD: Less changes, pkgsrc (also in dragonfly now thanks to GoBSD), probably more stable. To bad there are no nvidia drivers. Also had JAVA issues but I think they are solved now. (run Linux JAVA over their Linux “emulation” layer just as good as Linux does.)
Solaris 10: Would be fun to test, but I expect it to be slow, harder to get the regular open-source apps compiled and installed (pkgsrc exists for solaris aswell, don’t know if all the apps from in the netbsd “branch” of it compiles however). Also I suppose I wouldn’t have the nvidia drivers and I could totally forget about winex/cedega.
OpenBSD: Well, not this time, again. I’ve used it a lot earlier but NetBSD and FreeBSD makes much better desktops, also my 3.5 box crashes when I grep my irc logs, that would probably never happen in NetBSD. (I think there is a patch for that one, I have just been to lazy to apply it.)
Heres some LKML commentary on FreeBSD’s SMP implementation. As to why FreeBSD would model their SMP implementation on Solaris, I do not know.
http://www.kerneltraffic.org/kernel-traffic/kt20041019_278.html#3
You were great in Wild Things!
can somebody explain what is that “Giant Lock” and why it is so horible?