On Tue, May 13, 2008 at 7:58 AM, Mulyadi Santosa <mulyadi.santosa@xxxxxxxxx> wrote:
I have read this information in the mel gorman's PDF, but this is not something fixed and determined.
The system tries to maintain this proportion in the sizes of the lists.
Hence, there should be any policy that would try to be achieved.
That policy should be static (I think), even the lists LRU structures and dynamics, as you have said.
I concluded that this policy is to not increase the number of pages scanned when the policy of replacing pages is applied, not increasing the overhead. In my view, that's the only explanation to maintain this proportion in size between the LRU lists.
I din't understand well that part
I'm already reading this PDF. ;)
Hi Eduardo...
Good questions...
not only process owned pages, but kernel allocated ones such as
On Tue, May 13, 2008 at 9:02 AM, Eduardo Júnior <ihtraum18@xxxxxxxxx> wrote:
>
> Hi,
>
> I read that the lists LRU (active and inactive) store working set of the
> processes that are running.
buffer, socket buffer and so on AFAIK.
Hmmm? never measure it...but that's possible...even more I guess...
> I read that also on the active list is two-thirds of the inactive list.
depending on current real memory pressure..
I have read this information in the mel gorman's PDF, but this is not something fixed and determined.
The system tries to maintain this proportion in the sizes of the lists.
Hence, there should be any policy that would try to be achieved.
That policy should be static (I think), even the lists LRU structures and dynamics, as you have said.
I concluded that this policy is to not increase the number of pages scanned when the policy of replacing pages is applied, not increasing the overhead. In my view, that's the only explanation to maintain this proportion in size between the LRU lists.
not defined AFAIK. it's entirely a dynamic data structure.
> My source has not informed as the size of the inactive list was defined, if
> at compile time or runtime.
nope, dynamic..
> So, I wonder:
>
> - The size of the inactive list is defined that way?
> - The definition of the size of the inactive list has some dependence onnope...
> architecture?
AFAIK yes
> - The lists are initialized in the boot?
any kind of allocated pages..by allocated i mean the one which is
>Only pages of the image and
> structures of pages of the kernel are these lists? What kinds of pages they
> store?
allocated via kmalloc() and friends and not locked into RAM and must
exist in the entire life of the kernel.
I din't understand well that part
Sorry, none that I have. maybe mel gorman's PDF.
> Any reference to respect would be welcome.
I'm already reading this PDF. ;)
[]'s
--
Eduardo Júnior
GNU/Linux user #423272
:wq
