[LSF/MM ATTEND] Transparent huge pages: huge tmpfs

[LSF/MM ATTEND] Transparent huge pages: huge tmpfs

[Hugh Dickins]

I would like to attend LSF/MM this year; and most of all would like
to join Kirill Shutemov in his discussion of THP refcounting etc.

I admit that I have not yet studied his refcounting patchset, but
shall have done so by March.  I've been fully occupied these last
few months with an alternative approach to THPage cache, huge tmpfs:
starting from my belief that compound pages were ideal for hugetlbfs,
questionable for anonymous THP, completely unsuited to THPage cache.

We shall try to work out how much we have in common, and where to go
from there.

Huge tmpfs is currently implemented on Google's not-so-modern kernel.
I intend to port it to v3.19 and post before LSF; but if that ends up
like a night-before-the-conference dump of XXX patches, no, I'll spare
you and spend more time looking at other people's work instead.

I haven't checked through the other topic proposals yet: but once I'm
uprooted and there at the conference, expect to be able to engage with
them.  As before, I'd really like to spend some time in the FS room,
but shall find it hard to detach from MM.

Hugh

Re: [LSF/MM ATTEND] Transparent huge pages: huge tmpfs

[Kirill A. Shutemov]

On Thu, Jan 15, 2015 at 11:54:09PM -0800, Hugh Dickins wrote:
> I would like to attend LSF/MM this year; and most of all would like
> to join Kirill Shutemov in his discussion of THP refcounting etc.
> 
> I admit that I have not yet studied his refcounting patchset, but
> shall have done so by March.  I've been fully occupied these last
> few months with an alternative approach to THPage cache, huge tmpfs:
> starting from my belief that compound pages were ideal for hugetlbfs,
> questionable for anonymous THP, completely unsuited to THPage cache.
> 
> We shall try to work out how much we have in common, and where to go
> from there.
> 
> Huge tmpfs is currently implemented on Google's not-so-modern kernel.
> I intend to port it to v3.19 and post before LSF; but if that ends up
> like a night-before-the-conference dump of XXX patches, no, I'll spare
> you and spend more time looking at other people's work instead.

Very interesting! Looking forward for the code.

-- 
 Kirill A. Shutemov

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [LSF/MM ATTEND] Transparent huge pages: huge tmpfs

[Andrea Arcangeli]

Hello,

On Thu, Jan 15, 2015 at 11:54:09PM -0800, Hugh Dickins wrote:
> I would like to attend LSF/MM this year; and most of all would like
> to join Kirill Shutemov in his discussion of THP refcounting etc.

Seconded.

I'm positive about the tradeoffs coming from Kirill's simplification
of the THP refcounting by allowing to map compound pages in ptes (not
only in trans_huge_pmds) and in turn by relaxing the constraint that
split_huge_page cannot fail. It looks a worthwhile change even if it
spreads the complexity to a wider codebase and it doesn't keep it
localized in get_page/put_page anymore.

> I admit that I have not yet studied his refcounting patchset, but
> shall have done so by March.  I've been fully occupied these last
> few months with an alternative approach to THPage cache, huge tmpfs:
> starting from my belief that compound pages were ideal for hugetlbfs,
> questionable for anonymous THP, completely unsuited to THPage cache.
> 
> We shall try to work out how much we have in common, and where to go
> from there.
> 
> Huge tmpfs is currently implemented on Google's not-so-modern kernel.
> I intend to port it to v3.19 and post before LSF; but if that ends up
> like a night-before-the-conference dump of XXX patches, no, I'll spare
> you and spend more time looking at other people's work instead.

I'm interested the above topics too.

Thanks and hope to see you soon :)
Andrea

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>