OSNews

That's sweet. I remember a while back that fedora had wanted to replace the old init with something new. I'm glad to see that they finally decided on something.

I can't wait to try out FC9 on my SSD! New and improved, with KDE4! :-)

*EDIT* Yeah yeah, I know. Fedora 9.

Edited 2008-02-10 01:10 UTC

excellent news, would love to see fedora bootup a lot quicker

Interesting that Fedora is using technology initiated by Ubuntu..! (They are often the pioneer of new technology.)

Do you consider HAL, D-BUS, NetworkManager, Cairo, PulseAudio, Avahi and so on as Fedora technology anymore?

It is just upstream projects and there is bound to be cross pollination. Fedora did evaluate extensively all the options though. Refer

http://fedoraproject.org/wiki/FCNewInit

Of course you are right that it is usually the other way around but that is because Ubuntu doesn't usually initiate new upstream projects. Upstart being a marked exception.

I think the poster is referencing the recent onslaught of comments in every Ubuntu news post about how they don't innovate at all, and just integrate Fedora and Suse developed technologies. I wasn't going to be so nice with my smart ass comment, but I got beaten to the punch.

Popularity does that. I find that I can't read those jealous and spiteful comments without thinking of this old Morrissey song:

http://tinyurl.com/236clv

;-)

Even if that was true, I think this kind of integration is a contribution in itself. Finding ways to create a smooth-working system (be it server or desktop) from existing technology is a goal worth pursuing. And then there are bugs to fix, real-world scenario tryouts and so on. Implementation is just as important as creation, I believe.

Edited 2008-02-10 15:52 UTC

> Do you consider HAL, D-BUS, NetworkManager, Cairo, PulseAudio, Avahi and so on as Fedora technology anymore?
>
> It is just upstream projects

http://fedoraproject.org/wiki/RedHatContributions
cairo: written (employee) and maintained
dbus: written (employee) and maintained
hal: written (employee) and maintained
NetworkManager: written and maintained
Avahi: maintained
PulseAudio: maintained

> It is just upstream projects

Red Hat is always friendly with other distributions. So Red Hat try to put his projet "upstream".
For example libvirt/virt-manager are created by Red Hat :
http://libvirt.org/
http://virt-manager.org/

There are many other examples.

Great to see that we finally have a winner in the slow as as molasses drive to replace sysvinit. For years potential replacements have floated around, seemingly to no effect. With *buntu, Fedora, and subsequently RHEL and CentOS using upstart, I'd say the old init's days are numbered.

Fedora/CentOS/Ubuntu are my Linux workhorse trio. Between them, I can pretty well satisfy all the computing needs that I and my clients have. So it's cool to see good ideas flowing between those distros; That is how it should be.

But... I can't resist chiding superman about a comment he made just yesterday:

---

http://www.osnews.com/permalink?300101
---

So... who's riding on who's coat-tails now?

Fedora does important work, to be sure. But they don't have a monopoly on implementing good ideas. No distro does. And that's a wonderful thing.

Thinking about superman's statement, I couldn't help but be especially amused by the last sentence of the first link which Rahul kindly provided. :-P

Edited 2008-02-10 02:41 UTC

The deficiencies in sysvinit are not minor. They may have started out as small fractures, but the cracks are getting worse.

I do not see balkanization. Not in the Linux space, anyway. Initng and the others have been around for a long time, but aren't really used to any great degree... except by hobbyist dabblers, of course. ;-)

With *buntu already using Upstart, that's a big chunk of the Linux installations right there. But with Fedora now following, which implies RHEL and CentOS will, too, I'd say that we have critical mass in both the desktop and enterprise server spaces. Though it wouldn't surprise me if Novell dumped some new init system on us that they'd been working on internally, in secret, ala XGL.

Edited 2008-02-10 03:01 UTC

How big a chunk? How many installations? Where are you getting your numbers? And for the love of whatever God people freely choose to revere, please don't point to distrowatch.



You really think RHEL, and by implication CentOS, users care about boot times? Fedora does creative things, but it doesn't mean they'll all wind up in RHEL. RHEL is the opposite of Fedora, it wants stability. Give it five years of proven capability, and then maybe they'll switch.

Let's be serious. Just because Ubuntu farts, it doesn't mean everyone has gas.



Considering XGL and compiz were picked up by pretty much all of the major distros, including FC and Ubuntu, then maybe that wouldn't be a bad thing. I don't actually advocate the approach Novell took, but it's kind of a bad example to pick since it sort of worked out more successfully than anyone expected. AIGLX didn't see the light of day until Novell started pumping XGL, unless RH was working on it behind closed doors as well...

There are atleast two factual inaccuracies in your post. XGL has never been in Fedora and AIGLX has always been developed in the open from day one and became part of upstream Xorg release soon.

True on the XGL part, which is why I included compiz, which was also developed by Novell alongside XGL, and adopted by Fedora, but maybe I should have been more specific.

My point about AIGLX is that it was developed in response to XGL. They had talked about it for a long time, but it didn't actually gain traction and substance until Novell took the initiative, for better or worse. XGL was and remains a kludgy solution, even if Novell still refuses to admit it, but at least it got RH moving on it. Sort of like the way Ingo got around to creating a new scheduler only when it appeared that CK's might possibly get adopted. There's a bit of a pattern there, apparently RH is much like MS in that they become aggressively motivated when faced with competition to their choice technologies.

But that's all tangential, the real point was that XGL being developed behind closed doors still benefited the community, which in turn was frankly tangential from the original point of this thread anyways.

Elsewhere, you disappoint me. I'd have thought that you would be above this petty Ubuntu bashing and that state of denial that some people go into when evidence of Ubuntu's popularity is mentioned.

Now, you know very well that hard numbers on the use of any Linux distro are completely elusive. So your request for hard numbers is disingenuous, at best. However, I just spent a few minutes collecting some indicators of relative distro popularity. Start with this:

http://tinyurl.com/ywdd79

Make sure to take a good look at PCLinuxOS showing there, BTW, before proceeding on to this:

http://tinyurl.com/2kuwmc

( I comment further upon Distrowatch later )

And then have a look at the results of the 2007 desktoplinux.com survey:

http://desktoplinux.com/files/misc/2007-distributions-sm.jpg

Dell provides Ubuntu *exclusiely* as its desktop Linux choice, based upon user surveys done in preparation for their desktop Linux support. (I'll dig up a few links, if you want.)

And Prospective Lenovo customers certainly know what they want:

http://tinyurl.com/2hs6fa

I just now visited Walmart's site, and it appears to me that they have completely dropped their Microtel line of PCs, which ran Linspire and Xandros, in favor of the popular new Everex line running gOS... based upon Ubuntu.

And then have a look at a few statistics which I have just taken from some popular distros' public forums:

--
Ubuntu Forums:

Currently online: 7861 (671 members and 7190 guests)
Most users ever online was 12,818,
--

--
PCLinuxOS Forums:

Currently Online: 50 Guests, 20 Users
Most Online Ever: 511 (June 30, 2007, 01:47:52 PM)
--

-
Fedora Forums:

4940 (183 members and 4757 guests)
Most online Ever: 11706, 2006-10-30 at 03:36 PM PST.
--

--
OpenSuse Forums:

125 guests, 9 members 0 anonymous members
Most users ever online was 339 on Feb 6 2008, 01:02 AM
--

I share your distrust of distrowatch figures. However, the rest of the evidence aligns nicely with the figures there regarding Ubuntu. It's the PCLinuxOS numbers on DW which stink to high heavens when compared to other indicators.

Those are just a few indicators off the top of my head. I imagine I could dig up plenty more. It's not like they are hard to find.

So, now that I've shown you mine, you show me yours. Where are your numbers or indicators that Ubuntu does not represent a "significant chunk" (that was the phrase you contested in my post) of Linux installation? But at this point, I have to ask... are you a member of the Flat Earth Society, as well?

Now that I have dispensed with that silliness. Let's get on to the rest of your post:



Upstart is about much more than shortened boot times. In fact, the quicker boot, though much talked about, is really just a logical result of what Upstart is really about, and that is sane and intelligent handling of system services throughout the cycle from boot to shutdown. By mentioning only boot time, you betray your lack of familiarity with it and how it differs from the competition, which mainly focus upon parallelizing the boot process.

RedHat would most certainly be interested in intelligent management of services.

And I would argue that historically, a major feature's inclusion in Fedora is a strong indication of it's inclusion in subsequent RHEL releases. It might just make it into RHEL for 6.0 in sysvinit emulation mode, in 6 months to a year. Or it could be another 2-3 years, and show up in RHEL 7.0. Please explain how you obtain your 5 year estimate. I don't see any historical precedents to support it.

CentOS will do exactly as RedHat does, for obvious reasons.



I only highlight your comment above to note how it reveals the emotional, anti-Ubuntu motivations which apparently resulted in your writing the rest of your post. There are no other redeeming qualities of the statement to discuss.



Rahul has already pointed out the factual problems here. I will add that I did not mention Compiz. Only XGL, which was pretty much DOA due to poor decisions made by Novell as a result of not soliciting peer review on the design.

As to Compiz, Novell's approach resulted in one of the more prominent OSS project forks in recent times. A fork which resulted from the "cathedral", closed door, development style of Novell's Compiz project. So you see, my example is a good one. And would have been even if I had opted to mention Compiz, which I didn't but which you gratuitously interjected as a sort of intended straw man.

But all of the above aside. Why can't people just get over the fact that Ubuntu has become quite popular? Stop tilting at windmills ragging on Ubuntu and Canonical? Use the distro or distros which suit them best? And start simply enjoying the benefits of someone having done the work to bring Linux to the mainstream desktop in the way the Ubuntu and Canonical have?

Their work in that area, and on Upstart, benefits us all. It benefits me, even when I am using or proposing Fedora or CentOS.

Why can't people get over their irrational jealousy of Ubuntu's popularity and simply give credit where credit is due?

Like I say, Elsewhere, you disappointed me with your post, because I thought that you would be above such things.

One other note. Campaigning, unfairly, against Ubuntu is counterproductive even to the "Anti-Ubuntu" cause, because it prompts people like me, who enjoy multiple distros, and probably would not take the time to post much specifically about Ubuntu otherwise, to stand up and shoot you down with posts like this one.

Edited 2008-02-10 15:26 UTC

I'm not denying Ubuntu's popularity, but popularity and userbase are not the same. I get irked when people assume Ubuntu is representative of the linux userbase, and I don't single out Ubuntu in that regard. It's no different than the "Gnome is the enterprise desktop" or "rpm causes dependency hell" arguments that are more representative of group-think and assumptions, particularly the tendency to apply personal criteria against a group as a whole, without looking at the entire picture and rationalizing arguments. I'll call any argument out when I see it heading in that direction, even if I personally agree with it, because I don't think there is enough rational discourse. Too many discussions do get bogged down in emotional and subjective responses, I like numbers and objectivity.



Sure, I'm aware of that, it's why I brought it up.



You still haven't pointed to anything concrete, it's perceptions based upon the visibiility of Ubuntu and it's exhuberant userbase. The loudest group isn't always the largest group.

Taking Lenovo, for instance, it's Thinkpad line is predominantly business oriented. Are they surveying the correct market when asking what distro they would like to see? Does 14,000 users wanting to see Lenovo preinstall Ubuntu statistically translate into a realistic market potential for Lenovo?

Lenovo, BTW, does preinstall SLED on select models which makes sense to me considering their target market. Plus, Dell and HP will preinstall SLED or RHEL on their commercial desktops for their corporate customers when requested. It's just not something the make a check box option on the website.

Then there's the Eee PC, which has shipped in excess of 300,000 non-Ubuntu-based linux PC's. And they're forecasting over 3M this year, though I'd take that with a grain of salt. One could argue that the Eee PC has already given Xandros a more viable commercial marketshare than Canonical, since there are actual dollars attached to it, but I think it's more indicative of the fact that the DE in this type of scenario doesn't actually matter, as long as the system works well enough for it's intended purpose.

And as usual, all speculation of linux userbase ignores that little area of the world known as Asia. Last I saw, Turbolinux owned that market in terms of revenue marketshare, whereas Red Flag owns it (mostly due to their strength in China) in terms of install base. These are the companies that HP, IBM, Dell, Oracle et al. are working with. Novell is making inroads with strategic local partnerships, and Red Hat is making a concerted effort there as well.

I'm not in any more of a position to make proclamations about desktop userbase breakdowns. But my original point was that it's a big wide world out there, and the users that are active on forums and the blogosphere are one segment of a global userbase, and not necessarily representative of it. Yes, there is too little empirical data to draw assumptions from, but that doesn't make it right to draw assumptions from incomplete data either.



No, I'm familiar with, in fact I'm quite interested in it (yes, you probably would never have gathered from my post), but RH cannot contemplate moving from hairy old sysvinit until Upstart has really proven itself. That will take time.



Like mono did?

At any rate, RH is not aggressive with the core underlying system and doesn't change at a whim. Given that init is the mother of all processes and can impact the stability of the entire system, I can't see a platform focused on stability and reliability making that change until it has seen sufficient QA. As in enterprise-level QA, not enthusiast commmunity QA.

Fedora is the proving ground for such technologies, but it doesn't mean they will automatically flow up to the enterprise product.



Didn't argue that.



Please, it was a tongue-in-cheek statement. If we're at the point where asking rational questions about Ubuntu saturation equates automatically to emotional, anti-Ubuntu motivations, then the community has reacched a sorry state.



I brought up Compiz because bringing up XGL alone is half the story. Nobody will argue XGL sucks, though at least it provides a viable alternative for people unable to use AIGLX, and as I pointed out in my other response, XGL lead to AIGLX materializing in form rather than theory. Compiz was forked, which was not a bad thing. Novell's emphasis was stability over bling, the community wanted and got bling over stability. Now there's a little more sanity with both projects, and compiz has become widely adopted.

You made it a derogatory point that Novell developed XGL behind closed doors. I responded that it doesn't matter, because it wound up spurring innovation for all distros that we hadn't had up until that point. I'll let the poets argue about whether the end justifies the means.



All of the distros work to that area. Upstart is the first major project they've initiated and is drawing community interest, so it will be interesting to see how that develops. At any rate, where did I say anything to the contrary?



Again, where is that coming from? I haven't criticized Ubuntu, I haven't bashed it or slagged it, As I said, I don't single out Ubuntu here, I'll call out arguments based on assumptions and speculation any time I see them. You attempted to lay a foundation for your assertion, I still question it, but I would hardly call it an emotional anti-Ubuntu response.

However, in re-reading my thread, I can see now where I may have come off as appearing overly aggressive, and that's not my usual style, at least until it's warranted... so my apologies if you interpreted it that way. Wasn't my intent. No more posting for elsewhere after he's had a few too many Coronas.

Why wouldn't RHEL users care about boot times? Long boot times are even more annoying or even costly if you have perhaps hundred or so users waiting for the server to come up.

Then there is the feature of automagically restart services that unexpectedly died. That alone would be reason enough for these users to switch.

Also note that Solaris uses a similar system as upstart in its latest incarnations, so I find it very likely that RHEL/CentOS will use upstart once it has proven itself in Fedora for a while.

While your point is good, and one I've made about X11 many times, in the case of Upstart your derision is misplaced.

We are presently in the experimentation phase. Everyone agrees that sysvinit needs to be improved upon, but no one is sure of the best approach. What you have, then, is multiple people trying different things. Sooner or later the most useful ones will start winning and crowd out the less useful.

This is not a case like a text editor or even a desktop environment where swapping around is trivial or merely time consuming. You simply wont get lasting fragmentation in this area. I forsee at most two new init systems which will last alongside sysvinit and bsd-style init which--face it--are not going away this decade.

I like the approach taken with Upstart and see for it a potentially bright future, but only time will tell. I know it has some issues, mostly for me with regards to being sure of which process will start in what sequence (this can matter a lot, especially when diagnosing problems). The primary advantage upstart has today is that some distributions actually ship it by default which is more than can be said for most init replacements. That it was also thoughtfully designed bodes well.

We'll know in about five years which init system has won.

Er... how? There was no viable alternative at the time.

And there still isn't. For example, follow the InitNG link and note how, like the other candidates, it addresses parallelization of the boot process and not much else. SCO Unix was doing that in the 90s, for heaven's sake, albeit in a less sophisticated way. Upstart is the only candidate with dependency based, parallelized boot and shutdown, process monitoring, and event-based management of services even after boot up which can respond to hardware hotplug/unplug doing all the right things at the right time. All in one slick and elegant package. Not to mention taking over for creaky old Vixie Cron.

Upstart isn't just another InitNG clone. It's much more than that.

You make it sound like parellelized boot is a good thing. It is not. Parellelizing the boot process is a can of worms... there is no standardization on services' names, so it is difficult to specify dependencies. There is no critical mass of users using parallel boot, so upstream projects couldn't care less about it. Not having a strict initialization sequence makes it difficult to troubleshoot problems. Shall I go on...?

Server people don't care about boot times (heck, many high-end unix machines take 10-15 minutes just on the POST routine), and desktop users who care can just leave their machines in stand-by.

Should we start mimicking windows in its "Some services failed to start" messages too?

And for hardware hotplug we have udevd, what is the problem with that? Having it all in "one slick and elegant package" is just plain wrong.


What exactly is the problem with vixie cron? It works.

I mention it as one feature amoung several.

Upstart does serialized boot just as nicely, if desired by the distro, or the admin.

Distro vendors have full control over the configuration of their provided init systems. How do you think they decide on all those S01-S99 and K01-K99 prefixes that they use? So that's a pretty weak argument against parallelization.

Upstream providers don't have to. Linux distributors do.

Flip a switch and set the boot to run serialized for troubleshooting purposes. Or have that happen automatically if trouble is detected on a boot.

You haven't come up with one good argument yet. Keep shooting.

No server customers care about 5 nines? That's 5 minutes of down time per year. 26 seconds per day.

Those for which standby and wake-up actually work, that is.

I was wondering when you would throw in the "it's bad because Microsoft does it" red herring.

Hotplug and udevd are different things, altogether. They can be perverted into doing much of what is needed, in the same way as a dime can be used as a screwdriver, but why not use the right tool for the job?

You're going to have to support that with some specific arguments.

Use it to do something simple, like schedule something to happen every 20 days, and show me the result. That's just one example of its 1960s-1970s design.

Edited 2008-02-10 16:01 UTC

Just as a sidenote, FreeBSD's init system rc.d implements dependency based service startup since version 5 of the OS. Keywords in the rc.d/* file indicate needed preconditions and provide informations about which services or functionalities are available after service startup (PROVIDE, REQUIRE, BEFORE). System services are controlled via one central configuration file /etc/rc.conf. The rcorder mechanism takes care about the order which service to start. Shutdown works similar. Certain keywords (start, stop, restart, status etc.) can be passed directly to the startup scripts (e. g. /etc/rc.d/devfs restart).

As far as I see, this system does not address parallelizing startup. Event based actions are handled by the respective services (devd, usbd).

Regarding downtime because of system startup: While the system runs without problems, downtime isn't interesting at all. But in case of troubleshooting, it even does matter to server people - I know it, I get the complains. :-)

After all, I think Upstart may be a good choice because of the advantages you mentioned above. It is still important to have linear processing in case of troubleshooting. We'll see how average users and server people can benefit from this system, time will tell.

IIRC, dependency based sysv init is one of the release goals for Lenny. Upstart could be a possible goal for Lenny+1.

As a Debian user and fan I find that interesting. Do you have linkage?

I personally would not predict or expect upstart by default in Debian for quite some time, given how stodgy Debian tends to be. As an installable option, though, like initng is now, I like the idea.

No, I don't know any release goals for Lenny+1. I was just speculating.

Debian's new automatic init script generation infrastructure, called metainit, seems to be designed to make package migration as easy as possible between Debian (with sysvinit scripts) and Ubuntu (with upstart jobs). It will also make much easier for Debian to switch from sysv to upstart in the future, if upstart proves out to be good.
http://wiki.debian.org/MetaInit