How many is too many?

Posted Oct 13, 2005 3:55 UTC (Thu) by kirkengaard (subscriber, #15022) [Link]

Michael Dell wants less confusion so he can make an easy "right" choice. In the end, he must care more about his bottom line than about Free Software. His bottom line is the life of his business. The apparent fact that he is interested in Linux distributions may be the result of any number of factors; it may well be that Mr. Dell considers Linux a possible augmentation of his product line, and thereby an expansion of his bottom line. However, the question betrays a lack of Free Software understanding. The answer to "How many is too many?" is "Mu."

The profusion of distributions mirrors the philosophy of Free Software; GPLed code exists for everyone, and everyone has the right to use it to the fullest extent, provided that their users retain the identical rights.* For GNU/Linux, certain itches just have to be scratched with a customized system; the kernel and the tools must be combined to be useful. Some people, on the other hand, just have an itch to build a distribution. They all have the right to do so, and many responsibly give back to the community, which is also a great way to A) generate interest, B) get help supporting and testing, and C) see to the longevity of the created work. You may not need any of those hundreds of distributions; you probably only need a limited few of them. However, what you need in a Linux system depends largely on what you will be using that Linux system for. If a distribution exists that helps you scratch your itch, even if you must modify it to fully meet the need, Free Software has worked.

My concise answer is, "'As many as there are' is probably still less than the number of members of the problem set."

* Not a legally complete explanation of the GPL, I know. Go RTL.

Posted Oct 13, 2005 5:09 UTC (Thu) by jwb (guest, #15467) [Link]

Dear Mr. Dell:

All we want is hardware that works. Thanks.

-Linux Users

Posted Oct 13, 2005 6:29 UTC (Thu) by error27 (subscriber, #8346) [Link]

It turns out that it's not really as simple as saying if one distro works on the hardware then they all will.

For example, redHat and suse do a bunch of bug fixing on they're enterprise distro's. A couple major hardware vendors target the redhat suse kernels instead of kernel.org kernels. So even if the source is there, most people don't know how to port it to the most recent 2.6 kernel.

And the other thing is that it's not easy to update the kernel. For example recent FC3 kernels needed a newer kernel need a newer version of mkinitrd with a newer version of bash. There are selinux dependencies. Udev needs to be updated. I guess the HAL stuff needs to match as well probably. I'm not sure how the PCMCIA stuff all fits together but it probably needs to be updated as well.

There are wierd problems. I think that for a while Debian wasn't distributing the tg3 driver in their stock kernel because of licensing issues. That's a pretty important ethernet driver.

When you're talking about enterprise distributions those are often based on pretty old kernels and there can be a big difference in the hardware that they support.

Then there are a bunch of cases where the Linux has multiple drivers for the same hardware. I've seen cases where only the bcm5700 driver works and other cases where you have to use the tg3 driver. There are SCSI drivers where only the non working driver is distributed in the installer. There are other drivers where it's in Redhat the right driver loads but in Suse not only is the wrong driver loaded but it's compiled into the kernel so you can't change module load order and you can't install.

And then there are ordinary kernel bugs and regressions.

It's not always easy.

Posted Oct 13, 2005 13:37 UTC (Thu) by brugolsky (subscriber, #28) [Link]

C'mon, LWN isn't Slashdot. What possible relation could Michael Dell have to Linux? He sells sub-standard equipment (like the Intel EM64T) with pre-loaded operating systems from Microsoft. He cares not one iota for the values of choice inherent in free software; if you want vanilla, he's your guy.

[No offense meant to Matt Domsch and the other Linux techs at Dell, but they ought to understand this better than anyone.]

A long time ago, Dell sold Dell UNIX SVR4, and it was good: compiled with gcc, included free software, had excellent tech support, etc. But that was 1991.

Today their tech support is outsourced to some novice reading from a script.

Even buying a dual-CPU Windows XP workstation from them is fraught with hassle and delay. The BIOSes in their machines are notoriously bad. And it takes them an additional seven days to turn on RAID-1 mirroring in Windows XP!

Dell got rich selling vanilla. He might sell chocolate if you pester him. Strawberry is nowhere in sight.

Posted Oct 13, 2005 16:08 UTC (Thu) by southey (subscriber, #9466) [Link]

if Linux distros are defined by package management.

A certain company that Dell has very strong relationship with still sells at least four versions (98SE, 2K, XP and 2003) and embedded versions (CE), has related versions floating around (including 3, 95, 98, ME, NT) and DOS. Each of which are not fully compatible and not to forget the different service packs that some software require to run and others crash with. In many cases the different distros within the same package type are more like Windows service packs.

So perhaps we have too few distros?

Posted Oct 13, 2005 16:14 UTC (Thu) by martinfick (subscriber, #4455) [Link]

While support for common distributions would be nice, what hardware makers
should be providing is the drivers. First get the drivers out there so
that the distributions can package them for you if they want them. Then
if you think that they are not doing a good enough job of it, either work
with them and/or become a debian/... developer and package them
yourselves. Lastly, once these other things are done, providing
preinstalled linux would be greatly welcomed.

Posted Oct 13, 2005 16:49 UTC (Thu) by iabervon (subscriber, #722) [Link]

I think that the solution to his problem would really be a reference userspace for the kernel, which doesn't do anything particularly useful but checks whether the kernel is supporting all of the hardware and reports how it is doing so. Then Dell can test their machines with that, and tell distros what they have to do to run on Dell machines. Of course, this wouldn't guarantee that everything would work, because the distro could fail to provide a way to invoke the desired operation from userspace, but that's likely to be unrelated to the hardware configuration.