Robert Davidson [[EMAIL PROTECTED]] writes:
> I dont know much about compression and how it all works, so I was
> thinking that if the majority of the archive was the same, and say
> they only patched up one or two files, then the compressed data would
> be the same or very similar for the files that didn't change. That
> was the theory I was thinking on.. Am I wrong?
There's probably no single "right" answer, but since a compression
scheme that is working well tends to yield the most random data
patterns, a small change in input can create a very large change in
output. If you have a series of files that are fed into a compression
engine as a stream, changing one of those files will likely impact
everything that comes out of the engine past that point. And because
of the random nature of the compression output, it's less likely that
some period of data later, the image will start matching the previous
version.
How much changes in the overall image if only a small fraction of a
group of files that are compressed together are changed will depend on
how the files are grouped and whether the compression is being
performed file by file or on the entire grouping as a whole.
If you're taking a file at a time, compressing it and appending to
output (say a zip file), since things will reset at the next file, you
have a better chance of regaining similarity to a prior version at
some point. Of course, if the individual file is also large, the
compression will result in changes throughout the file itself beyond
the point of any changes. But if you are tar'ing up all the files,
and then gzip'ing the tar image, you're probably going to find the
compressed image significantly different past the point in the
compression where the changed files were encountered.
-- David
/-----------------------------------------------------------------------\
\ David Bolen \ E-mail: [EMAIL PROTECTED] /
| FitLinxx, Inc. \ Phone: (203) 708-5192 |
/ 860 Canal Street, Stamford, CT 06902 \ Fax: (203) 316-5150 \
\-----------------------------------------------------------------------/
