Hi,

I have played with the page scanner and finally merged the two inactive
queues into one.
I think this is important after the integration of the ubc code as with
distinct queues we will have the following situation:
1. invocation of the scanner: scan the swapbacked objects until free
target met
2. invocation of the scanner: scan the vnode backed objects until free
targtet met
and so on

this means if there are a lot more vnode pages in the system (what
happens from ubc) then swapbacked ones this behavior is quiet
"unbalanced" as the "smaller queue" is scanned too much.

After merging the two queues the system became much "smoother"...
I am still working on this...
Probably it would be a good idea to be a bit more aggressive about
cleaning pages on the inactive queue even if they are not paged out....

lars