Linus Torvalds explains why the unexpected resilience of kernel version 2.6 has delayed the move to kernel version 2.7. In this two minute video he said that when work started on 2.6, he was worried that major changes would destabilise the kernel.
Linus Torvalds explains why the unexpected resilience of kernel version 2.6 has delayed the move to kernel version 2.7. In this two minute video he said that when work started on 2.6, he was worried that major changes would destabilise the kernel.
Well i guess the kernel has reached that stage were it has to remain stable to continue progressing, however i do miss the old days when we would see crazy updates being added every now and then, however this way lots of time can be spent on further work and no extra time on backporting.
What linus is saying is that they’re being able to drop “crazy” features on every kernel release.
So I don’t know why you “miss” the old days when crazy stuff was being added “every now and then”. Take a look at http://kernelnewbies.org/Linux26Changes and look at the features that are being added…there’re “crazy” features, big modifications on how the VM works…etc etc. Just because it’s not news it doesn’t means it doesn’t exists.
I believe that there will eventually be a new stable kernel series in the not-so-distant future. Not because of any major architecture overhaul or anything like that, but because there will be demand for a slower-moving kernel. This demand exists today, but it just isn’t practical yet.
The way I see it, the parallel development of the past was necessary to move the kernel ahead without leaving non-kernel-hackers in the dust. The 2.6.x kernel series represents the first time where the overall maturity of the kernel makes it possible to develop the kernel without crapping on the user-base. Maturity doesn’t necessarily equal stability, but the kernel project really can’t worry about that right now.
The 2.6.x series is the big push towards the rich feature set demanded by the desktop and server markets alike. Once we get closer to 10% market share, we won’t be able to make aggressive changes as easily as we can now. It’s now or never.
Linus has said that one of the major reasons he doesn’t want to open a 2.7.x development branch is that he would have half as many testers. Certainly if the kernel had more commercial users and distributors, there would be more testers. But there are a lot of features that Linux didn’t have in the 2.4.x series or in 2.6.0 that prevented its commercial uptake.
There are a number of key kernel features in the pipe that must be completed to address these markets. Namely, it needs standard facilities for full, para, and container virtualization. It needs enhanced serviceability features such as checkpointing, tracing, and (more) useful crash-dumps. It needs an overhauled VFS that provides pooled logical volumes and an enterprise-class COW filesystem to compete with ZFS.
Once we have these in place and widely tested on the 2.6.x kernel series, I would like to see a new kernel series that brings more stability and longer release cycles. A true enterprise-class Linux kernel. I’d say it would deserve the 3.0 moniker.
Edited 2007-01-17 16:00
Although a bit short, it’s still a good video. Nice to see that Linus does not obsess about the details
Speaking of which, the guy does seem very down to earth.
But it needs flash 8, so we can’t read it under Linux.
and in the other news, flash 9 final has been released for linux
@IceCubed Modded you down by mistake
flash is a pain under Linux. I would rather have a working gnash.
I agree with you on gnash. I wish it would be more feature complete.
On topic:
I just remembered the 2.4.x and 2.5.x trees, and now look at 2.6.x and I’m kinda surprised to see the 2.6.x as a stable development (oxymoron?) kernel.
Linus always has been down to earth. Look at the DRM issues and the upcoming GPLv3 stuff.
He’s pretty clear on those. I wish more people were like that; it wouldn’t cause so much damage to linux and it’s community as seen lately by the overheated reactions on the MS/Novell deal.
The only damage done to Linux was by Novell.
the oly damage done is by people like you who firmly believe that interoperability is a bad thing. If you think a it further, you also understand that the patents clause/issue is of no use at all. What’s left is the opportunity to blend in a heterogenous network, something that’s seriusly wanted and needed in the real world.
GPLv3, FSF and the community that don’t understand it or don’t _want_ to understand it, are causing the damage.
Look at the reactions from Linus — he basically says the same.
You’d better stop feeding the FUD, you’re helping MS. It’s your call…..
the oly damage done is by people like you who firmly believe that interoperability is a bad thing.
the oly damage done is by people like you who firmly believe that microsoft getting all patenty and fake-interoperability-y is a good thing. The number one barrier to interoperability with Microsoft has always been…microsoft, and if they have changed, where are the gpl’ed microsoft linux kernel patches to make samba work flawlessly?
If you think a it further, you also understand that the patents clause/issue is of no use at all. What’s left is the opportunity to blend in a heterogenous network, something that’s seriusly wanted and needed in the real world.
If you think a it further, you also understand that the patents clause/issue is a serious barrier to true open systems. What’s it left us is the opportunity to once more obfuscate the world into umpteen incompatible networks, something that’s seriusly unwanted and uneeded in the real world.
GPLv3, FSF and the community that don’t understand it or don’t _want_ to understand it, are causing the damage.
Proprietary and DRM-loving companies and the community that don’t understand it or don’t _want_ to understand it, are causing the damage.
Look at the reactions from Linus — he basically says the same.
Look at the reactions from Linus — he basically says the exact opposite.
(viz. that DRM is A Bad Idea.)
You’d better stop feeding the FUD, you’re helping MS. It’s your call…..
Yes, thankyou for helping MS.
Being down to earth is good for engineering but hardly is the last word in other realms, especially when it means “I don’t care.” I hope no one’s head explodes over this, but LT and other kernel developers are fallible, and just a little of the insight on this comes from a former* Novell employee:
http://tieguy.org/blog/2006/09/22/what-the-kernel-guys-got-wrong/
(*) Lest anyone get their panties in a bunch, note that Luis amicably moved on from Novell well before the MS/Novell deal.
Down to earth should never involve apathy, it just involves keeping your emotions in check so you can deal with problems and ideas in a more factually correct manner.
I know plenty of really down-to-earth people who are in fact, quite passionate about things (not necessarily technology), without being going over the deep-end emotionally. Emotions tend to cloud logical thinking. You shouldn’t abandon emotions, conversely, you shouldn’t cave to them either.
Humans are funny things. Some are just funnier than others.
Down to earth should never involve apathy, it just involves keeping your emotions in check so you can deal with problems and ideas in a more factually correct manner.
Excessive emotions are a problem but a side issue here.
From The Collaborative International Dictionary of English v.0.48 [gcide]:
down-to-earth down-to-earth adj.
facing reality squarely; guided by practical experience and observation rather than theory. Opposite of {idealistic}, {unrealistic}, {impractical}, and {pie-in-the-sky}.
The original post implies that down-to-earth is preferable to, say, idealism. This flies in the face of great people in history who were previously disparaged as idealists.
To remain down-to-earth, LT must avoid talking about freedom. Freedom is, in a sense, an inconvenient truth.
Edited 2007-01-17 17:37
“Humans are funny things. Some are just funnier than others.”
Yeah, those humans never learn.
I don’t really care about this ‘damage’. If others want to use GPL’d software, then they should not try and circumvent the goal and spirit a la Tivio. If that means telling corporate business where to shove it, so be it. They have to learn to play by the rules and spirit of the GPL. DRM does not conform with the spirit of the GPL. If you don’t believe that, you’re no friend of the GPL or open source software.
Nor do I consider Linus a true friend to the GPL for obvious reasons – he’s more interested in corporate interest than the ideals of the GPL. I wish that the Linux kernel had NEVER been released under the GPL to be honest.
Dave
Kernel, right now is so stable and almost features complete; what is not is the GUI on top of it. Why I am saying this?
Well, I do ask myslef this, when was the last time I saw a kernel panic or a complete OS freeze? never in the past 4 years; and when asking when was the last time GNOME or KDE restarted due to a crash, I would say 2 hours ago.
Of course all of my experiences were obtained from 6 heavily used system at my home that runs redhat 4.4 AS with latest updates installed. So if one system is a deviation from the curve of logic why all systems keep doing it over and over…so I concluded that the GUI is the source of trouble.
Alot of problems in linux experiences are not coming from the kernel but the bugginess of the projects that sets on top of it. Add to that also the weakness of features of OSS software and the OS tools and you will really salute the kernel for being the only stable thing in Linux/GNU combo.
Good luck though GNU and I am looking for the day you could let me erase all broken windows without hesitation.
Well, I do ask myslef this, when was the last time I saw a kernel panic or a complete OS freeze? never in the past 4 years; and when asking when was the last time GNOME or KDE restarted due to a crash, I would say 2 hours ago.
Ask the microkernel guys to take their ideas to the userspace world and fix that, they’re so obsessed with kernels that they don’t realize that the main source of unstability today is userspace, not kernels! A bug in the CDDB library crashes rythmbox (or it was serpentine) everytime i put a CD that doesn’t have a CDDB entry. A bug in the kopete webcam feature crashes the whole kopete. Userspace is a interesting place to practice isolation between features of an app, kernels are a small part of the game these days!
Edited 2007-01-17 12:23
indeed, userspace is much more unstable. but crashes aren’t that disastrous either – most apps now have autosave (which will be improved in KDE4) so you won’t lose (much) data, and they can be restarted quick and easy, esp with functions like the crash plugin in konqueror (lately firefox has something like that as well). An Xserver crash is pretty serious, tough, but they don’t happen very often either.
it’s not like i’m arguing app crashes aren’t bad, of course, they should be prevented as much as possible, but you must agree it’s not as bad as a kernel crash…
A bit on a tangent…
I find features like autosave a really odd solution to crashing problems. It’s effectively saying that “We don’t care that our program crashes, we don’t care to fix it. Or actually we don’t know how. Instead, let’s add this autosave feature to not offend users.”
Well, it’s pretty much as you already said.
well, after all – you can’t realistically eliminate all bugs… so autosave is at least helping
I refer the interested reader to Dave Jones’ amusing and enlightening paper from the 2006 Ottowa Linux Symposium, entitled, “Why Userspace Sucks.”
https://ols2006.108.redhat.com/reprints/jones-reprint.pdf
This is more about performance and power management than about stability. You can make stat() as fast as you can, but it doesn’t matter if applications stat thousands of files repeatedly for no good reason.
You can make stat() as fast as you can, but it doesn’t matter if applications stat thousands of files repeatedly for no good reason.
Well – if you want to f.e. sort a directory with the subdirs first, you pretty much have to call an fstat on each entry. Other methods are not POSIX compliant and even depend on the kernel number.
Well – if you want to f.e. sort a directory with the subdirs first, you pretty much have to call an fstat on each entry. Other methods are not POSIX compliant and even depend on the kernel number.
But you only notice a drop in performance if you do in the same thread as the GUI…
I think stat() is a bit of a bad example here, because it involves uerspace *and* kernelspace. Especially, you can speed it up by adjusting the kernel API (e.g. allow batch stat()-ing). I think the previous poster meant to apply uK ideas to pure userspace parts.
What video driver are you using? In my experience that’s the number one cause of crashes in Linux GUI apps.
“kernel, right now is so stable and almost features complete; what is not is the GUI on top of it. Why I am saying this?
Well, I do ask myself this, when was the last time I saw a kernel panic or a complete OS freeze? never in the past 4 years; and when asking when was the last time GNOME or KDE restarted due to a crash, I would say 2 hours ago. ”
That’s because GUI are one order of magnitude more complex than kernels, in a way. Writing a kernel is difficult, but the interface (from an API point of view) is quite small, and the algorithms well defined. Also, the user never uses directly the kernel as he does use those applications.
For example, the interface between the linux kernel and the application on the top of it is what, a few hundreds functions maximum. Now, a basic graphical application uses at least ten times more functions, with often richer data, input by users, etc…
So the complexity is quite different. Kernels are difficult to write, no question about that, but in a way which is finally more manageable than complex GUI applications.
Good luck though GNU and I am looking for the day you could let me erase all broken windows without hesitation.
I can’t possibly begin to imagine the confusion of ideas that resulted in this statement.
Your problem almost certainly lies in one of GNOME/KDE, Xorg, and/or your graphics driver, none of which are in any way related to the GNU project.
Your problem almost certainly lies in one of GNOME/KDE, Xorg, and/or your graphics driver, none of which are in any way related to the GNU project.
From the “About GNOME” web page:
“GNOME is Free Software and part of the GNU project…”
Maybe in the beginning, but GNOME is no longer a real part of the GNU project in anything other than name. Let’s put it this way: GNU doesn’t oversee GNOME any more than it oversees KDE.
In the last 8 months when I started using Gnome again, it has never crashed. I’ve had KDE crash though, even though I seldom use it. But running Redhat you must have lots of experience with Gnome (crashing, I guess).
How the hell does this comment get modded up? GNU’s userland software causing stability issues? That’s bloody bullshit. Your comments are FUD, and should be modded down as such, sadly it’s not an option when using the moderation system. osnews.com staff can we add a FUD option to the moderation system?
The kernel isn’t perfect, and certainly has its problems, much as any other software does.
Dave
Off-topic
=========
It must be me one of the most influential men in open-source speaks out, and we have grunts that ApplicationX doesn’t work. Mod these people down.
I say this for my own sanity not others. I turn into help desk mode and what Linus says gets drowned out.
I’m going to address all your user-space concerns. I’m pushed to think of a single application on Linux that doesn’t have multiple replacements. Report the bug. Use a different one. Join a forum, go on IRC ask for help. That is appropriate.
I’d have 2.6 bronzed and mounted, sucker is golden.
Totally ridiculous load of crap you can’t watch the video in linux without a proprietary binary plug-in.
I’m trying to watch the videos from my Windows laptop, and I can’t even get it to play either.
Of course, if I right-click anywhere in my start menu hierarchy, Explorer crashes and restarts, so this may not be the best Windows install ever.
Hmm… no proprietary plugins for me, and I can watch video just fine. Only flash is supported by a binary plugin (but flash isn’t really video though) and then of course one binary proprietary driver (nVidia).
But all audio and video codecs are GPL’ed on my Gentoo system, and it works fine (except for DVD playback.. gstreamer and Totem sucks in that regard).
My understanding is that the video *is* Flash, like YouTube and other embedded video content on the web. So your proprietary plugin is hard at work rendering content to your proprietary display driver (which is by itself about the same size as the rest of your kernel image). So much for free software…
Lucky me I’m not particularly religious about licenses – and lucky me I don’t consider flash valid for anything but Joe Cartoon.
Most embedded video content do not make use of flash – and YouTube isn’t even worth mentioning.
And there is much to free software. OpenOffice, Firefox, Thunderbird, Mono, Java, Linux (kernel), Gnome, KDE, bzflag (can’t live without it), Wine, Lazarus, FPC, Gaim, eclipse, and so on. The fact I have room for Skype, nVidia, flash and America’s Army doesn’t invalidate or somehow reduce the value of open source/free software.
But all audio and video codecs are GPL’ed on my Gentoo system, and it works fine (except for DVD playback.. gstreamer and Totem sucks in that regard).
Try the xine backend (it’s called totem-xine under Debian/Ubuntu, not sure about Gentoo), it works much better.
I’m aware of xine. Haven’t tried it though, but I could always change my Use flags. gentoo doesn’t have a specific totem-xine package due to it’s source-based nature
surprising that he is surprised about that…
is this why patrick volkerding((slackware grandmaster)) wants to continue using 2.4????
-2501
Part of the reason some folks prefer 2.4 over 2.6 is the fact that some older hardware isn’t as well supported by the newer kernel.
just like newer hardware isn’t supported by 2.4… so it depends on what hardware you aim your distribution.
That’s why *both* kernels are available (one 2.6 in /extra, another in /testing, and some 2.4 in /a).
You can always compile your own kernel anyway
What he describes as resilience could just as easily be explained by programmer complacency. The same 50 or so core programmers have been hacking at the kernel for years, it seems they have it pretty much as they like it now. That doesn’t mean there aren’t scores of programmers who deride the kernel for its bloat and sloppy design. There are. That they have decided to turn away any suggested sweeping changes is comforting on one hand, but a little upsetting on the other. I’ll feel better when 2.7 launches if it ever does. The longer complaints go unaddressed, the more likely a fork is.
It’s interesting how good his english has become and how he seem alot more like a people person now, sorry but that was the first thing I thought when I watched that video.
He’s from Finland, you can’t get much more down to earth then that,
As far as the kernel goes, can’t say much because I’m one of those odd persons with hardware that the kernel just seems to love, everything always just works for me.
I agree is English is superb and his peoples skill rock, the guy is well spoken.
This is no gripe with Fins, in fact quite the contrary, but the Fins have always seemed very wacky and out there to me or at least all the Fins that I have met. The sheer amount of Scandis here in Luxembourg is quit impressive and the Fins in particular always seem to have what it takes to turn a dull moment into lots of laughs.
Just my €0.2