Re: Interest in DOC and YAFFS? --> YAFFS bootloading

[Marc Singer <elf@buici.com>]

On Tue, Sep 24, 2002 at 10:30:45AM -0700, Russ Dill wrote:
> > > > A question, though.  I've been doing compression tests with cramfs.
> > > > I'm finding that gzip -9 of an ext2 filesystem produces smaller images
> > > > than mkcramfs.  Have you ever compared the two?
> > > 
> > > cramfs is meant to be lean, fast, and low on ram consumption, if you
> > > compress the whole thing at once, you have to load the whole thing into
> > > ram to read any of it, so cramfs compresses PAGE_CACHE (4096) sized
> > > pages at a time
> > 
> > That's what isn't clear.  I made two filesystems with the same
> > contents.  One cramfs and the other ext2.  The ext2 filesystem
> > compressed was smaller than the cramfs.  My understanding is that both
> > must be uncompressed into a ramfs to be used.  If this is correct,
> > then the only comparable consideration is the size of the compressed
> > data.
> 
> no, a cramfs does not need to be loaded into a ramfs, only the pages
> that are needed are loaded from the cramfs, and if memory is in a pinch,
> fs pages can be dropped. If you gzip a 4M file at once, vs gzip 4096
> byte pieces of it at a time, the former will end up smaller. (deflate
> uses repetition of information, and runs of things).
> 
> of course, it depends which you want, greatly optimized memory usage
> (cramfs), or a slightly smaller image.

Yes, that is the trade off.  I was unclear about how cramfs was
loaded.  It was my understanding that the cramfs was unpacked into a
ramfs so that the fs could be used R/W.

I'll evaluate it again when I have a chance.