On Mon, 14 Oct 2002, Sean 'Captain Napalm' Conner wrote:
It was thus said that the Great Tothwolf once stated:
I have 486 boxen running both 2.2 and 2.4
kernels, which seem to work
much more efficiently than the ancient 2.0 kernels. As far as
distributions go, Debian seems to be better suited for installs on
small hard drives, but I managed an install of RedHat on a 120MB drive
once (and I swore I'd never do it again).
I tried compiling a 2.4 kernel for one of my systems (static
compilation, no modules---I tend to forgoe modules for servers) and it
was (with the same settings) about twice the size as the 2.0 kernel I'm
currently running. My 486 systems are a bit tight with memory so that
is a concern for me.
I did forget to mention something quite important about the 2.2 kernels
when running them on older hardware. Kernels of the 2.2 series prior to
2.2.19 or so have a tendency to act *very* slow on these systems. I never
could track the problem down, but I think it was something to do with IRQ
or DMA handling. The problem did not show up on 586 and later hardware,
and it still puzzles me. I had no trouble at all once I installed 2.2.20
and later.
Did you happen to notice how much memory the kernel may have freed at
bootup? I've not noticed a significant increase in terms of memory usage
from 2.2 to 2.4. Both 2.2 and 2.4 can be compiled much
smaller than 2.0,
if you disable everything you don't really need. Both versions
are much
preferred over 2.0 when used for embedded applications...
The 2.2 and
2.4 Linux kernels support IRQ sharing, so it should work
"OK" so long as you don't attempt to run the shared ports at very high
baud rates.
What about the hardware?
Some cards handle it better than others. I only had a few cards that
didn't want to get along. Strangely enough, for older cards, I had the
best luck with cards that had real 16550A uarts on them instead of custom
multifunction I/O chips. Many of the newer multi-I/O cards can cope with
shared IRQs too.
-Toth