> Here is another attempt to address the "couldn't get a free page" effect..
> I think that the basic problem during "bonnie" is:
Heureka, It works!
I applied your patch to a clean pre6 (= I 'patch -p0 -R'ed Krzysztof's
and Benjamin's patches). The 8th bonnie run is going and no messages
so far. I knew it would work when It started the second bonnie because
the machine didn't get jerky at all and swapping was barely hearable.
Wohooaau! Even two simultaneous bonnies won't choke it. Load is < 2
and the machine is incredibly usable (nonpatched kernel was jerky).
When it was time for the seeks, load jumped to 7.5 but the machine
still stayed very usable.
Oops, I'll better jump back to my pants :)
-- Samuli Kaski, samkaski@cs.helsinki.fi Department of Computer Science, University of Helsinki, Finland.