Why are there only i686 and i586 Version of glibc and kernel? -- not "lite" (386/486 @ 400MHz+, 256MB RAM)

[Per Bjornsson]

On Sat, 2004-06-05 at 19:54, William M. Quarles wrote:
> Alan Cox wrote:
> > Nobody would argue much about 486, but 386 really hurts and I've seen
> > no modern embedded core that isnt 486 ISA
> 
> OK, especially since a Red Hat developer has said this, then why has Red 
> Hat Linux and Fedora Core remained 386 architecture for so many years? 
> Why not make it 486 architecture with 686 (or better) scheduling?

Well, first of all the compiler flags used are AFAIK -march=i386
-mcpu=i686 so the scheduling is already i686 - it's really not as bad as
is sometimes implied by slashdot trolls and others.

One reason why the move from i386 to i486 hasn't been made might simply
be that it _hasn't_ actually hurt much earlier. It seems that the main
benefit of ignoring i386 is that one can link explicitly with NPTL,
which is of benefit for threaded applications. This is pretty new
though, NPTL was introduced in Red Hat Linux 9 IIRC, and the suggestion
from Jakub Jelinek was that from Fedora Core 3 all Fedora Core
applications should be built explicitly against the NPTL libraries
instead of using the "legacy" threading model interfaces to glibc. What
I believe Alan means with "really hurts" is that using the new NPTL
interfaces can be a real boost for threaded applications, and i386 lacks
some features (available in i486 and later) that are essentially needed
for NPTL. Thus this has only now become an important issue. (And it's
really quite possible that the architecture for most packages will stay
"i386" even for FC3 although because there's no NPTL port available for
i386 it will be impossible to run the distribution on a 386 without
using in-kernel i486 emulation or something like that.)

/Per
(Not speaking for anyone, any misunderstandings are probably mine)

-- 
Per Bjornsson <perbj at stanford.edu>
Ph.D. Candidate, Department of Applied Physics, Stanford University