On Monday 26 November 2007, Jeff Spaleta wrote: > On Nov 26, 2007 3:21 PM, Lamar Owen <lowen@xxxxxxxx> wrote: > > I've followed Les's argument for a long time; he (and others) want kernel > > stability but not userland stability. > How about this, how about you stop attempting to interpret Les's > comments and just speak for yourself. Ok, fine. There are machines where I would like this as well. And I have it with CentOS plus KDE-Redhat. KDE-Redhat is much more than an add-on; it is more of a 'Fedora Alternatives' for CentOS or other RHEL-sources based distribution. Remember Fedora Alternatives? Ask Rex about the packaging; I'm just thankful he did the packaging so that something more modern than KDE 3.3 can run on CentOS 4. Thank you, Rex! (The version of kstars that can control telescopes is post-3.3; I actually NEEDED kstars of recent vintage running quite a long time prior to the CentOS 5 release; see my .sig for details). This doesn't help GNOME users, though. > I understand the point you are > making concerning ABI stability and you clearly understand that > upstream kernel development doesn't value it. It's absolutely not > clear to me that the lack of ABI stability is something Les really > cares about nor understands why its difficult for Fedora to provide. Not speaking for Les, but I understand why it is difficult for Fedora to provide this (backporting is expensive; ask the PostgreSQL group about that; if it weren't for Red Hat and RHEL, security fixes probably wouldn't be backported to older PostgreSQL versions); but where else can the issues that this difficulty produces for Fedora be discussed? LKML? LOL. > > Why? If this were true, kde-redhat's RHEL repo wouldn't need to exist. > > KDE-Redhat for RHEL exists, therefore this isn't true. > It doesn't "NEED" to exist... its "DESIRED" but you "CAN" live without > it. Fedora doesn't NEED to exist... it's DESIRED, but you CAN live without it. Lather, rinse, and repeat with various other Linux distribution names. The fact that packagers have put effort into a monster job like KDE-Redhat (or PlanetCCRMA) speaks volumes for the perceived need. I already mentioned my real NEED for kstars for KDE > 3.3. > It's much easier to provide a fast moving application stack as an > addon to a conservative distribution then it is to build a > distribution that is conservative and fast moving in different areas. Rex is on this list; Rex, would you characterize this as being true (since I should only speak for myself, and not try to interpret what others have already clearly said in other venues as well as this one)? Rex is the KDE-Redhat expert. > What was the point of this thread again? The point, as I see it, is that Fedora will eventually break your system if you follow kernel updates. The corollary is that the developers don't care as long as the political 'no stable module ABI' point is being made. The breakage may be minor, and it may be major. I'm at F8 on this laptop; when it was an F7 laptop I used it for audio production (I do professional multitrack production as a side item); some F7 update broke the sound. Completely. I had wireless mostly working. Some F7 update broke it, too, and renumbered my ethernet. Note that the wireless and the sound drivers were open-source modules bundled with the kernel, and not binary drivers. I'm not even going to talk about what it did to how my Tascam US428 USB DAW control surface/sound interface worked. And the tickless kernel threw VMware into a tizzy (part of my audio production requirements at the time meant I had to use Adobe Audition, which I used under VMware; with the tickless kernel audition became unusable because the audio edit cursor jumps around and no longer tracks the audio that is playing). (The nohz kernel command line option might have fixed that, but I didn't know about that at the time) Now, I needed recent Audacity, Ardour, and other programs to do the work; it was a kernel issue, not a userland one, that the sound card simply would not produce any sound and the US428 just quit being recognized. In the interim, I went back to doing my production on the F7 desktop with the Echo Layla; less convenient, but when production is at stake, one makes sacrifices. Almost enough to make me want to throw Fedora out the window and buy Studio2Go. Almost. > If its come down to ABI stability in the > kernel, then we are at an impasse. Just because there is an impasse doesn't mean that users who despise this political motivation in kernel development should just shut up. I think the kernel developers (and perhaps the Fedora developers) should hear what users think about the system in this area. Users want to use the system, and they don't want it to break mysteriously due to one CRITICAL-FOR-SECURITY update that also happens to include a new 'feature' that breaks what was working. I too enjoy some of the new features in the 2.6 kernel series (like in-kernel ivtv drivers (no firmware, but that's easily available), or echo audio card drivers). But I also subscribe to the 'do no harm' principle of development; a change should not be made inside a major version that causes serious regressions for users of the system. Or do we need to consider 2.6.22 -> 2.6.23 a major version update? If that's the case, then, IMHO, Fedora shouldn't update the kernel major version inside a Fedora major version. It's not too hard to wait six months for the newer feature to come out. If that's too long, track rawhide (that would really put the relative stability of Fedora in a new light, as it IS fairly stable for the most part, in the released versions). As it is the stated goal of Fedora to stay on the edge, so to speak, I know better than to put a Fedora box into production as a server. And that's part of the point. Fedora is a great (and I think it is the best) Linux desktop for enthusiasts. (Yes, I have installed and run Ubuntu 7.10, done the PCLinuxOS thing, tasted Mint, done the SuSE jig, etc; Fedora meets my needs the best of the bunch on the desktop). I would not dare use it as a server. Too fragile when it comes to updates. If you never had to update, sure, it's a great server OS. Just don't connect it to the Internet without updating. CentOS or RHEL is a much better choice here. I would not dare put it into nontechnical users' hands as an enterprise desktop, either. I have enough user service requests to deal with without dealing with the regressions introduced by a newer kernel version. CentOS or RHEL plus KDE-RedHat fills that niche for my uses. If Fedora wants to be a server-grade distribution an enterprise desktop distribution, then kernel module ABI stability should be a goal. If Fedora doesn't want either of these, and wants to ignore the module ABI problems, then, Ok, that's fine. Just don't misrepresent Fedora's suitability for those uses. If it's a bleeding edge enthusiast's desktop, then bill it as such. Fedora can be rock solid (it is on every box I've put it on). Rock solid running isn't the issue (for me!); having a nonzero number of regressions after a kernel update is. The OP's point was trying to debunk the idea that Fedora is just a beta testbed for RHEL (of course, the point of the thread has morphed since then). Playing the devil's advocate for a just a moment, ask yourself this question: would it be in Red Hat's interest for Fedora to really be update-stable on either the enterprise desktop or server? Red Hat does an enormous amount of open source development work that goes unrecognized; the Fedora distribution is part of that (PostgreSQL is another), and I for one am thankful and grateful for all Red Hat does. And I don't think Red Hat intentionally 'cripples' Fedora, even though it may appear that they do. -- Lamar Owen Chief Information Officer Pisgah Astronomical Research Institute 1 PARI Drive Rosman, NC 28772 (828)862-5554 www.pari.edu
