Replication

Multiple copies of an archive give better guarantees of safety from loss, but as each copy costs storage, a replication strategy is important to ensure a balance of safety and economic sustainability over time.

Preservation best practices recommend a "3-2-1" model: 3 copies of the data in at least 2 different supports, 1 of which off-site. This should protect against a variety of threats, from hardware failures to ransomware attacks, to natural disasters.

As for digitization quality and primary storage choices, this too comes down to the archive's budget. In a tight situation, if one has a primary store on RAID, it may be acceptable to only have a second copy off site; or less expensive options can be considered for the secondary copies, such as tape and deep (cold) storage. The latter (e.g. Amazon Glacier) costs a small fraction of regular cloud storage, but one must know what these savings come with: retrieving data back from deep storage can be quite expensive and take very long times. Therefore, a deep archive copy can be used as long as it is expected to be never accessed, except in an extreme disaster case where all other replicas are lost.

Replicated stores are not controlled by Pocket Archive. To automatically propagate updates from the primary store to the replicas a number of tools can be used, for example rsync or Borg. These tools can be set up to run, e.g. nightly to synchronize the replicas with the source. rsync can perform differential backups (to only copy the changed files) and has many powerful options, and Borg is capable of keeping a fixed number of previous versions that rotate out, if one can afford the space.

[TODO add recipe for basic replication setup with rsync and/or Borg]