Currently we are considering only the userspace
anonymous pages for movement.
As of yet, we have not considered kernel pages for
movement due to the 3GB rule.
Since userspace allocations are by page-faults, we
consider only single pages
(0th order) for movement. Thus, only the pages which
are not file-backed & are 
on LRU lists are currently being considered for
movement.

We now want to broaden our definition of a movable
page, & consider kernel
pages & file-backed pages also for movement. Do
file-backed pages also obey the
3GB rule? In order to move such pages, we will have to
patch macros like "virt_to_phys"
& other related macros, so that the address
translation for pages moved by us will take place
vmalloc style, i.e., via page tables, instead of
direct +-3GB. Is it worth introducing such
an overhead for address translation (vmalloc does
it!)? If no, then is there another
way out, or is it better to stick to our current
definition of a movable page?
Identifying pages moved by us may involve introducing
a new page-flag. A new page-flag 
for per-cpu pages would be great, since we have to
traverse the per-cpu hot & cold lists
in order to identify if a page is on the pcp lists. 

As of now, we have adopted a failure based approach,
i.e, we defragment only when 
a higher order allocation failure has taken place
(just before kswapd starts swapping). 
We now want to defragment based on thresholds kept for
each allocation order. 
Instead of a daemon kicking in on a threshold 
violation (as proposed by Mr. Daniel Phillips), we
intend to capture idle cpu cycles
by inserting a new process just above the idle
process. Now, when we are scheduled,
we are sure that the cpu is idle, & this is when we
check for threshold violation & defragment.
One problem with this would be when to reschedule
ourselves (allow our preemption)? 
We do not want the memory state to change beneath us,
so right now we are not 
allowing our preemption.
This will ofcourse hog the cpu, & we may not be able
to reschedule just by checking
the need_resched flag. Any advice or suggestions
regarding this problem? Also, will
the idle cpu approach be better or will the daemon
based approach be better?

Any suggestions,advice & comments will be highly
appreciated. 

Thanking you,
-Alok


__________________________________
Do you Yahoo!?
Yahoo! SiteBuilder - Free web site building tool. Try it!
http://webhosting.yahoo.com/ps/sb/