On Thu, 22 Mar 2001, Alan Cox wrote:

quoted

pull it all right back in again.  It continues through the entire
build with the cost seen in the time numbers.  (the ac20.virgin run
was worse by 30 secs than average, but that doesn't matter much)

Using my reference interactive test (An application which renders 3D graphics
and generates fairly measurable VM traffic with AGP texture mapping)[1] the
graphical flow is noticably stalling where it didn't before.

Throughput seems to be up but interactivity is bad.

If you set the amount that kswapd goes after to be a fraction of
inactive_target and leave Stephens change in but ensure that a
schedule happens between loops, IIRC interactive is pretty nice
while swapping.  (haven't tried that particular variant in a while)

	-Mike

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/