Avi,
Thanks for your response.
Regarding the query on "Guest Swapping", I'm referring to the mechanism that
the KVM host can use to swap out a guest's pages. Since a guest OS will have
its own swapping mechanism, then how will the host ensure that if it choses
to swap out a guest's page, it will not conflict with the guest's own
swapping mechanism.
Put another way: I understand (from docs, mailing list etc.) that KVM allows
the host to overbook/overcommit physical memory via the guest swapping
mechanism. i.e. the guest thinks it has x number of physical pages but
actually the host has swapped out some of the guests pages and given the
freed up physical memory to another guest VM. So the guest actually has less
than x pages in physical memory but doesn't know it.
It possible that my understanding of the way KVM works may be incorrect, in
that case any clarifications would be appreciated.
Thanks again
Arjun
On Sun, Apr 6, 2008 at 3:39 AM, Avi Kivity <[EMAIL PROTECTED]> wrote:
> Arjun wrote:
>
> > Hi Folks,
> >
> > A fellow student and I wish to do run some experiments with KVM.
> > Specifically, we would like to examine
> > KVM's guest paging/swapping mechanism, make some changes and run some
> > tests. After a brief search through
> > the docs and code, we would greatly appreciate help with the following:
> >
> > 1) Guest Paging : What is the selection mechanism used by KVM for
> > evicting a guest's page ?
> >
>
> With mmu notifiers (not yet merged), pages are picked by the normal Linux
> LRU mechanism.
>
> What if there are any conflicts between host and guest as to which page
> > to evict ?
> >
> >
> I don't understand this question.
>
> 2) Where can we find the code for the above (guest paging).
> >
>
> It's the normal Linux swapping code.
>
>
> > 3) Are the KVM guests scheduled by the (hosts) standard linux scheduler
> > ? It appears that this
> > is the case but I want to be sure.
> >
>
> Yes.
>
>
> > 4) Have there been any changes to the host's page/buffer cache ? Are
> > there any changes planned ?
> >
> >
> Yes, mmu notifiers. This doesn't change pagecache, but rather allows
> kvm's shadow mmu to be synchronized to the main Linux mmu, which is what
> allows page eviction and page aging to work.
>
> --
> error compiling committee.c: too many arguments to function
>
>
-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Register now and save $200. Hurry, offer ends at 11:59 p.m.,
Monday, April 7! Use priority code J8TLD2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
kvm-devel mailing list
kvm-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/kvm-devel
