When mirroring a directory containing versioned files with rsync, it
seems that extra data is being transferred unnecessarily when a new
version of the file is put in the directory.
For instance, a directory contains files:
someprog-3.7-beta1.ext
otherprog-9.2.ext
...
and this is mirrored to another system with rsync. Then the site
updates someprog-3.7-beta1.ext to someprog-3.7-beta2.ext (or other
similar name) and deletes the someprog-3.7-beta1.ext . If you are
rsyncing with the --delete option, the old file will be scheduled to be
deleted and the new one copied over in its place without doing a
difference test between what are really two similar versions of the same
program.
Could rsync compare the list of files to be deleted and the ones that
will be added, and match files being deleted that are "similar enough"
to the new files being created (filename starts/ends with the same
characters, size is within 10%, replacement has a newer date, etc), and
then use that as a basis for the download block comparison?
Currently if you want to do this (for instance with large .iso files),
you have to manually check to see what the new name would be, rename the
current file that you have, and then run rsync, in order to see savings
in the amount of data transferred.
Just a suggestion. Keep up the good work!
Alvin
