On Thu, Mar 10, 2005 at 06:46:05AM +0900, YAMAMOTO Takashi wrote:
> hi,
> 
> > At memory shortage condition, sum > 100% makes the page daemon
> > abandon page-access-history due to the page-queue-reordering effect.
> > That's one of things that I'd like to avoid.
> 
> i don't think tweaking sysctl is a right way to avoid it.

let me clarify that as "changing the usage-balancing tuning settings
doesn't seem to be the best way to improve the access-history mechanism".

yea, I agree.


> it reminds me an old patch in my local tree.  (attached)
> i suspended it because i don't think tuning page balancing etc makes
> much sense without fixing PR/27030.

for folks reading along, PR 27030 says:

        when doing pageout, uvm don't care about pages in kernel_object.
        while they're counted as anon, the most of them are wired and
        can't be paged out.  it imposes unfair penalty upon other anon pages,
        and make pagedaemon unnecessarily busy.


this is what the yamt-km branch is all about, right?

-Chuck