Fedora is starting to use s3cmd sync to mirror package repositories into S3. Fedora master mirrors make extensive use of hardlinks to reduce data size and data transfer from the master mirrors to downstream mirrors. S3 does not have any concept of a hard link, and s3cmd sync uploads the same (hardlinked) data multiple times, which is inefficient.
I would like to see s3cmd sync learn about local hard links, and issue S3 COPY commands after the first upload, for subsequent hardlink targets of the same file, to reduce upload bandwidth and time consumed considerably.
This is in my copy-hardlinks branch now.
merged into 1.5.0-alpha1.