> > how to better scale towards the very high end,
> > we will want to solidify those before 2.3 is started)
>
> Alas, I think, kernel developers should neither forget about
> so-called low-end systems, like 32M ram P66 or 5x86.
We won't, but the current algorithms are mostly based
on the assumption that CPU is plentyfull and I/O is
expensive.
This breaks horrendously on very lange boxes, where
system overheads of 40-50% are common.
We'll have to assume that both CPU and I/O are expensive.
This will mean that the selection code for pageouts will
get somewhat simpler. The negative effects of that can
be offset by doing better I/O clustering and swapin readahead.
> Don't forget, the point is, that linux makes a perfect Xterminal of
> a 386 with 8M ram (and should further versions do so too).
It probably will. It might even work better :)
> How far did 2.1.x kernels got in self-fine-tuning mem/swap/buffer/etc.
> parameters to fit a certain mem/proc./architecture config they are run on?
> Could it be improved by eg. some serious surveys on how a config performs
> with different parameters?
I don't know, but since I'm working on a performance
tuning document anyway, I might just as well ask folks
to do some tuning themselves and report 'dangerous' and
extremely-good settings to me.
Rik -- typing slowly because my kbd is dvorak since sun 19:40...
+-------------------------------------------------------------------+
| Linux memory management tour guide. H.H.vanRiel@phys.uu.nl |
| Scouting Vries cubscout leader. http://www.phys.uu.nl/~riel/ |
+-------------------------------------------------------------------+
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/