[OLPC-devel] Re: Announce: Compressed cache 'alpha-002' release

[Nitin Gupta]

The patch is against 2.6.18-rc3-git7 which is (I think) same as
2.6.18-rc4. So, please apply it against -rc4 not -rc3.

(little) testing was done on x86 _only_ (taken RAM=128M,
max_anon_cc_size=1000 pages -- started KDE, launched apps like KMail,
firefox, kpaint etc.)


On 8/7/06, Nitin Gupta <nitingupta.mail at gmail.com> wrote:
> ------------
> Project: Compressed Caching for Linux
> Git (web): http://dev.laptop.org/git.do?p=projects/linux-mm-cc
> Git (git): git://dev.laptop.org/projects/linux-mm-cc
> ------------
>
> Hi!
>
> This is weekly work summary [10/12] or I thought, better call it 'Announce' ;)
>
> Attached is patch for compressed caching for anon pages (2.6.18-rc3) -
> 'alpha-002' :)
>
> Finally, 'better' locking scheme in find_get_page() is working together
> with compression structure. This equates to 'nearly' complete compressed
> caching feature for anonymous pages. But, I haven't yet made any benchmarks
> though!
>
> Many of the suggestions from previous post for cache-alpha-1 have been taken to
> this patch. I will be glad if you can take time to look at this patch too for
> _any_ of suggestions you have. There are also some known glitches which I
> couldn't solve now.
>
> Known problems:
> -- Pages once decompressed are not put back on LRU (active?). Doing so is
> causing system freeze and I couldn't get to the cause yet.
>
> -- LZO is still giving problems. System freezes just after some pages are
> compressed. Biggest problem is that it requires mammoth 64K buffer to compress a
> single page.
>
> -- How to 'migrate' incompressible pages to swap disk? This is not an issue for
> OLPC as they do not have a swap.
>
> So, here's what I aim by end of SoC:
> (Riel, can you please suggest here?)
> * Either
>         Start work on to add support for page cache pages and heuristic for
>         adaptive ccache resize.
> * Or
>         Work to stabilize this much for anonymous pages, code cleanups, testing
>         on OLPC board, (maybe) document all this stuff.
>
>
> Cheers,
> Nitin
>
>
>