On July 28, 2002 07:15, Thomas Zander wrote:
> Last I heard file buffers are still the first to go when memory is needed.
> Even well before swapping (data OR code) is done in the first place. Did
> you misunderstand Werner, or what?

Code pages and the VM allocator pages are both file-backed buffers [1]. You 
might know the difference between something that's been mmap()'ed to store 
data, and something that's been mmap()'ed to be code. However, all the kernel 
sees is an executable bit set, and it really doesn't care about its state 
when it comes to the page replacement algorithim.

And no, currently nothing is swapped preferentially in Linux. The least 
accessed pages go first. It doesn't matter if it's file backed or anonymous. 
The only benefit I can see is if people haven't allocated enough swap space, 
and we're trying to outsmart the VM by giving "anonymous" pages an alternate 
backing store.

-Ryan

[1] Well, not really buffers. I assume you meant page cache, as buffers are 
pretty much only used for pending I/O in Linux, and caching block device 
access in 2.4.
_______________________________________________
koffice-devel mailing list
koffice-devel@mail.kde.org
http://mail.kde.org/mailman/listinfo/koffice-devel