[Beowulf] scheduler policy design

[Mark Hahn]

> There is a unique numbering system for linux kernels?  What is it? I've
> always thought of kernels as being part of distros.

sacrilege!

distros are just eyecandy, and the kernel they ship is normally 
patched out of all recognition.  sometimes some suffixes are added
to hint of this, but they understate the case: 2.6.9-42.0.10.ELsmp
is only vaguely related to the true 2.6.9 from kernel.org.  I don't
know whether distros have caught onto the fact that kernel.org 
has a nice system for choosing sequences of stable branches within 
the sequence of development versions (2.6.20.9 for instance - there
is no stable version of 2.6.21 yet.)  here's the last few kernels:

-rw-rw-r--    1 536      536      40956221 Apr 22 23:06 linux-2.6.16.49.tar.bz2
-rw-rw-r--    1 536      536      43377437 Apr 25 21:21 linux-2.6.20.8.tar.bz2
-rw-rw-r--    1 536      536      43997476 Apr 26 03:23 linux-2.6.21.tar.bz2
-rw-rw-r--    1 536      536      43373745 Apr 26 07:15 linux-2.6.20.9.tar.bz2

note two active stable branches...

in any case, kernels are a thing unto themselves, certainly not beholden
or subsurvient to distros.  I normally use kernel.org kernels on machines 
where I care (servers, clusters), but lazily leave the distro kernel on my
desktops.