Tagged APT snapshots' backup is impractical

For each release we currently add about 6-7GB of data to backup, which is painful when running backups on a poor Internet connection. I think we should investigate deduplication:

• either in the source filesystem itself, which has the advantage of saving storage space on lizard;
  • using hardlinks-based deduplication tools) should work, e.g. http://jak-linux.org/projects/hardlink/ that we use during our ISO build process
  • using a filesystem that deduplicates data would not help on the backup side (unless we use tools specific to that filesystem to back up our data); and last time I checked, no such filesystem was ready for production use on Linux
• or in the backup process itself, e.g. using bup instead of rdiff-backup
  • bup supports pull-style backups (see bup-on(1))

Related issues

• Blocks #13242

Original created by @intrigeri on 11830 (Redmine)