[Beowulf] scheduler policy design

[Mark Hahn]

> It's great that all y'all recognize a reliable standard source of good
> kernels, but I'm missing something.

all kernels come from kernel.org; the most current is always from there.
but kernel.org doesn't make any real claims about testing - distros do,
but it makes their kernels lag, sometimes by more than a few years.
to make up for this, distros often backport patches from more modern kernels.
to a lesser extent, distros also include features that are not in kernel.org.

> My understanding of the license is that
> I can download Red Hat, hack and slash to my heart's content (e.g.
> integrating symbolic galois extension fields into tcsh), and sell it as
> "Pete's Linux Version 1.00" and the kernel would just be whatever I
> compiled to boot ab initio.
> No?

you can download RH sources, and produce your own version of them or 
the binary products.  you can't sell it as RH only because of TM,
but Centos is very nearly just RH sources compiled again.

you can hack as you wish, including on purely user-level stuff like tcsh.
you could redistribute your hacked tcsh, or with Centos, or with your own
I-recompiled-compiled-from-RH-srpms distro.  RH is probably not the origin
of tcsh, so you could even make a My-Centos-but-newer-tcsh if you were 
really unhappy about critical tcsh features (hah) that hadn't been picked
up by "the upstream distro" (as Centos phrases it.)

anyway, just to be explicit, years ago, stable kernels where named 
major.minor.release with minor even (development/unstable kernels
had minor odd.)  that stopped a year or two ago, when modern source/patch
control systems made it possible to treat versioning much nicer.
now there is just 2.6.* for the mainline/development stream, and occasional
forks of 2.6.*.release "stable" branches (where 'release' can go pretty 
high, such as 2.6.16.49: forked from the dev 2.6.16 kernel, but with 
49 updates since.)