Re: [lc-devel] Compressed Caching on 2.6

[Anderson B. <bri...@gm...>]

On 5/29/06, Nitin Gupta <nit...@gm...> wrote:
> Hi,
>
> >
> >> >
> >> > About CC, I and Briglia (and...@gm...
> >> > <mailto:and...@gm...>) are making some tests with
> >> > Rodrigo Castro version (kernel 2.4) for ARM platform. We are studyin=
g
> >> > a possible port for 2.6 and would like to cooperate with project.
> >> >
> >>
> >> Great!
> >>
> >> My exams (and other misc things) just ended today...now I can really
> >> jump for compressed caching :)
> >> Are you doing ccaching for "general" desktop systems or some kind of
> >> embedded device?
> >
> > Some kind of embedded device. :) Actually we tested and did some
> > measurements using the Rodrigo's patches for kernel 2.4.18. We tried
> > to implement a swap compression on 2.6 kernel, but we stoped when
> > recording to disk presented itself as a problem.
> >
> (I assume you are using flash disks for storage).
> So, have you dropped the idea of swap to flash disks? (Compressed)
> Swapping to flash disks have lots of issues as far as I have read about i=
t.

We were using a mmc card as swap device. I believe its behavior is
similar from an IDE block device, for instance. The problem was
related to the bio structure: how to send more than one page if the
bio includes only one page per operation?

Best regards,

Anderson Briglia

>
> >> Just in case you didn't notice it earlier -- I have made a page
> >> linux-mm.org/CompressedCaching giving all the details collected by now
> >> for ccaching implementation on 2.6.x kernels (CompressedCaching/Code h=
as
> >> the bare bones implementation created by now). Also, this is where I'l=
l
> >> keep track of the project.
> >
> > Yes, we have been visiting your project's site. Will you use just the
> > site or we should communicate by this list as well?
> >
> The site is just to keep design details updated. We can always
> communicate through this list (and #linux-mm-cc).
> >>
> >> I intend to now focus on ccaching w.r.t OLPC systems. Thus, I'll now
> >> work on handling anon pages first instead of page cache pages.
> >>
> >> The biggest thing is compression storage structure -- heart of all thi=
s
> >> work. I've actually received 0 feedback on this by now. Since that
> >> involves my semi-random guess work too, it really needs your help :)
> >> All other work revolves around this thing. In particular please see
> >> 'Problems' listed at end of that page.
> >
> > Ok. We intend help you and the others to design this crucial part.
> Thanks!
> > How about that chunks approach? We thought you already decided to use i=
t.
> Yes, I've decided to use this chunk approach as described on
> CompressedCaching page. But, still if you can give your feedback, we can
> always improve it.
> >>
> >> PS: I've 0 idea of ARM systems!
> >
> > Don't worry about this. The kernel core is similar. Of course we've
> > other issues to care about, but the design of ccaching can be used on
> > both systems.
> >
>
> Best Regards,
> Nitin Gupta
>
>