Nitin Gupta wrote:
> Hi All,
>
> Just took some days off from compressed cache. Back to work now :)
>
>
Welcome back then :)
> Also, current implementation is going towards 'chunked' structure of
> ccache. It almost seems like a joke that without any simulations or
> detailed study I'm so inclined to implement it instead of earlier
> '2-page cell structure' which was so thoroughly studied.
> I think that after the initial implementation it shouldn't be too
> difficult to implement both (or maybe others) of these storage schemes
> for proper benchmarks.
>
I read a paper "Adaptive Main Memory Compression" by Irina Chihaia and
Thomas Gross, where they implement something they call "zones" to handle
memory. In short, what they do is the same as you proposed earlier
(linking fragments of compressed pages), but with an added locality of a
zone, so that you can increase and decrease the size of the cache by
adding or removing a zone (all fragments must be within one zone, that
can be comprised of more than one page). Just to comment on their
approach, they focus a lot on large datasets applications and only them,
IMHO we can do as well or better on large datasets and still don't throw
away performance for smaller ones..
> Please post any ideas you have for ccache.
>
I have not been to much help as far, as I'm still reading up on
everything I need (and writing a summary of it), but I will try to get
some of my ideas out there :).
Mvh,
Asbj=F8rn Sannes
|
