JFFS2 ram memory usage

JFFS2 ram memory usage

[Øyvind Harboe]

I've made some changes to JFFS2 locally to reduce memory usage.

The change I did was to disable compresssion as I have lots
of flash and little ram(Atmel EB40a 2MB flash + 256k ram +
eCos and my app).

My usage of JFFS2 is intermittant and hence I'd like to
reduce peak memory usage(much better with compression
disabled) and idle memory usage(perhaps unmounting JFFS2
when I'm not using it?).

I did post a RFC on the changes I did to the ecos-patches
mailing list, but I'll clean it up to the best of my abilities
if there might be useful to others as well.

Comments?

http://sources.redhat.com/ml/ecos-patches/2003-08/msg00019.html

Sincerely,
Øyvind

Re: JFFS2 ram memory usage

[David Woodhouse]

On Thu, 2003-08-07 at 10:42, Øyvind Harboe wrote:
> I've made some changes to JFFS2 locally to reduce memory usage.
> 
> The change I did was to disable compresssion as I have lots
> of flash and little ram(Atmel EB40a 2MB flash + 256k ram +
> eCos and my app).

You could also remove the actual compression routines if you're going to
do that. Other users, especially eCos users, may want to individually
select the compression algorithms available too.

> My usage of JFFS2 is intermittant and hence I'd like to
> reduce peak memory usage(much better with compression
> disabled) and idle memory usage(perhaps unmounting JFFS2
> when I'm not using it?).

Finding a way to eliminate the 'totlen' field of the jffs2_raw_node_ref
structure might be productive. Or possibly making the jffs2_raw_node_ref
contain doubly-linked lists if that then means we can discard obsolete
nodes immediately rather than keeping them around.

-- 
dwmw2

RE: JFFS2 ram memory usage

[Øyvind Harboe]

>You could also remove the actual compression routines if 
>you're going to do that. 

In my patch I:

- removed the allocation of the compression/decompression buffer
that the compression routines uses to attempt to compress.
- removed the invocation of the compression/decompression call.

>Other users, especially eCos users, may want to individually 
>select the compression algorithms available too.

Agreed.

I haven't checked, but I can't imagine that the runlength
compressor using a lot of memory.

>Finding a way to eliminate the 'totlen' field of the 
>jffs2_raw_node_ref structure might be productive. Or possibly 
>making the jffs2_raw_node_ref contain doubly-linked lists if 
>that then means we can discard obsolete nodes immediately 
>rather than keeping them around.

I don't speak enough JFFS2 currently to comfortably attempt this.

Øyvind