Hi Rik,
Here what the main concern is that the pages that are more closely located on disk should be evicted first and the randomly allocated blocks should be given privelege to be in cache.
Can we use some reverse mapping kind of mechanism that is used when the pages are reclaimed?
I am not sure about it..
Kindly give your comments or suggestions...
-Adheer
On 9/17/06, Rik van Riel <riel@xxxxxxxxxxx> wrote:
adheer chandravanshi wrote:
> Currently only temporal locality(number of times page is accessed) is
> used in maintaining the page cache.
> But we can also use the spatial locality(here: sequentiality of pages on
> the disk).
That makes sense. Worth trying out...
> To check if the page is accessed in the recent time we can have a
> timestamp..
> But when the pages are in the LRU lists(inactive list) how can we check
> the sequentiality of the pages?
This is hard, since pages in the page cache are indexed by
(mapping, offset), not by disk layout.
However, if you assume that filesystems have good enough
algorithms to prevent fragmentation most of the time, the
two could be close enough to indicate linearity on disk.
--
What is important? What you want to be true, or what is true?
