broomberg wrote:
RH AS is not for business. Hell, it's FREE on the web.
You know, I believe I posted earlier a claim that RHAS included some non-redistributable components, but now I'm not sure. RH's Web site is completely unenlightening on the subject, but I gather from elsewhere that all the contents are lawful to redistribute. I hear that one can gather all the SRPMs. Thus, if you wish, as a non-customer, you can do that, build as many as that will build right away, boot the partial system, then rebuild the rest. In theory, you could then construct ISOs, and distribute those. Nobody bothers to do that, but I see no reason they couldn't. (If I were a customer, I could check the product details for any proprietary claims to shipped binary CD contents. At a distance, I can't easily tell.)
My recollection is that almost all of the differences traditionally lay in RHAS's "enterprise" kernel RPM (which is always available in source form, as required by the licence given the binary's distribution to outside parties). That was where RH offered many of its patches first, such as their version of the "bigmem" patch to malloc up to 4 GB RAM on i386 -- but their "enterprise" kernels were only a little less dodgy than their regular ones, until the day they hired most of VA Linux's laid-off software engineering department. Now, at least their NFS and eepro100 support isn't as unreliable, but I'm still not convinced their kernels are particularly good.
It would beinsane to maintain support for 5 years of Linux "consumer" releases. There can be a release every 6 months. And if RH doesn't, SUSE, Mandrake, Debian, etc, etc, will. So RH HAS to.
The way things are shaping up, one thing you'll need to pay significant dinero for on a periodic basis is a guaranteed stable application interface for proprietary binary userspace software. You may recall that RH's unsubtle moves with the so-called "gcc 2.96" were directed towards hastening stabilisation of support for C++ -derived binary modules. That was all about shrink-wrapped software.
People (including lwn.net) bitch from time to time about the Debian Project "ending support" (updates) for numbered versions (Debian 2.1/slink, 2.2/potato, 3.0/woody...), and I always thought those complaints were Unclear on the Concept of that distribution, in that, if you just keep doing "apt-get update; apt-get dist-upgrade" as intended, you'll progress smoothly through stable=slink to stable=potato to stable=woody to stable=sarge, etc., and at every step is "supported" in the intended sense. But, as reliable as that is, it may or may not stably support binary-only software you want to run on it. That's part of what RH offers for money -- and SuSE, which for practical purposes is the sole substance behind United Linux.
The real long-term threat to RH dominance in this area isn't SuSE/United Linux, nor IBM's well-intentioned but mostly clumsy and ineffectual arm-twisting, by the way, but rather the LSB. Which, by coincidence, I'm hoping to give some help to, soon (technical writing and such). Not that I personally care a lot about guaranteed support for proprietary Linux i386 binaries, but it's a worthwhile task to re-level that playing field.
Rick Moen
rick@linuxmafia.com