On Wed, Nov 14, 2012 at 03:40:37PM -0800, David Rientjes wrote:
> On Wed, 7 Nov 2012, Kirill A. Shutemov wrote:
>
> > From: "Kirill A. Shutemov" <kirill.shutemov@xxxxxxxxxxxxxxx>
> >
> > H. Peter Anvin doesn't like huge zero page which sticks in memory forever
> > after the first allocation. Here's implementation of lockless refcounting
> > for huge zero page.
> >
> > We have two basic primitives: {get,put}_huge_zero_page(). They
> > manipulate reference counter.
> >
> > If counter is 0, get_huge_zero_page() allocates a new huge page and
> > takes two references: one for caller and one for shrinker. We free the
> > page only in shrinker callback if counter is 1 (only shrinker has the
> > reference).
> >
> > put_huge_zero_page() only decrements counter. Counter is never zero
> > in put_huge_zero_page() since shrinker holds on reference.
> >
> > Freeing huge zero page in shrinker callback helps to avoid frequent
> > allocate-free.
> >
> > Refcounting has cost. On 4 socket machine I observe ~1% slowdown on
> > parallel (40 processes) read page faulting comparing to lazy huge page
> > allocation. I think it's pretty reasonable for synthetic benchmark.
> >
>
> Eek, this is disappointing that we need to check a refcount before
> referencing the zero huge page
No we don't. It's parallel *read* page fault benchmark meaning we
map/unmap huge zero page all the time. So it's pure synthetic test to show
refcounting overhead.
If we see only 1% overhead on the synthetic test we will not see it in
real world workloads.
> and it obviously shows in your benchmark
> (which I consider 1% to be significant given the alternative is 2MB of
> memory for a system where thp was enabled to be on). I think it would be
> much better to simply allocate and reference the zero huge page locklessly
> when thp is enabled to be either "madvise" or "always", i.e. allocate it
> when enabled.
--
Kirill A. Shutemov
Attachment:
signature.asc
Description: Digital signature