Robert Collins wrote:
On Mon, 2004-10-25 at 13:04 +0200, [EMAIL PROTECTED] wrote:
I too used to think that too large swap area doesn't hurt,
but the explenation I heard, which sounds reasonable to me
though I never got around to test it, is that "lots of swap
also means lots of book-keeping" -
the kernel takes more time to manage the free swap pages and
scan through them.
Of course if you reduce your swap to a level that your application
can't handle your data then it's a problem.
IIRC there used to be problems with bad-O algorithms in the page table,
but in 2.6 they are fixed as part of the scalability work. Certainly
there is extra overhead in non-swappable kernel memory in maintaining
the larger page table. Thats fairly minimal (~32 bytes per 4Kb on ix86,
IIRC).
The overhead I though about was more about the CPU time required
to sift through the records rather than the amount of memory it takes.
I haven't heard about the 2.6 improvments you talk about, so what
you are saying is that "large swap is not such a bad idea" under 2.6?
Cheers,
--Amos
--
SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/
Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
