Brandt, Mark H wrote:
IMHO it would be more efficient to get IBM and Novel to address the
issue and eliminate  VDISK memory creep without having to configure
multiple VDISKS with different priorities.
That would fundamentally change the way linux memory management works
today:
Pages are distinguished in "anonymous" and "non anonymous" ones. Those
that are non anonymous have a backing on disk, for example they belong
to /bin/bash when running your command interpreter, or to a library.
Anonymous ones usually come to life by malloc(). They don't have a
backing on disk.
When time comes that a page gets to the end of the inactive list,
which means it is about to be discarded from main memory, we check if
the page is anonymous or not. If is non anonymous (that is, page has a
backing on disk) we write back the changes and discard it.
For anonymous pages (no disk backing), they get a swap slot assigned
from the lowest prio swap available and are considered non anonymous
from here on: they're written back to their swap slot and discarded.
When that page gets accessed again later, it is considered to be swap
cache with a backing on disk: the swap slot.
No method is in place to ever change a backing. Thus, a page will
never be moved from one swap slot to a different swap slot. This is
why linux is not very good at keeping swap usage local, and is not
good at using low prio swap for hot pages.

btw: I'd like to recommend Mel Gorman's book on this. Pdf is free on
the internet. The title is "understanding the linux virtual memory
manager".

cheers,
Carsten

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to