by This list is written from the perspective of a smallish research lab I support and is by no means exhaustive. However, it seems a good starting point. Feel free to share your ideas as followup posts.
Pros to tape:
– Can be easily taken offsite
– no spinning platters to jam
– more likely to last 4+ years under different storage conditions
– can’t become a “shared resource” (half backup, half ‘live’ storage)
– Well established backup technology. We know how bacula works and what it can do.
– Easily handles all three classes of backup (fumble-finger, archive, offsite/disaster recovery)
– lower power & heat requirements
– losing any quantity of tapes does not reduce the usefulness of any remaining tape
– Can keep adding as many tapes as you need, the drive is a fixed cost
Cons to tape:
– Slightly higher cost/GB _for the media_, amortized over more storage the cost can become lower than d2d
– Can’t be re-purposed as a file server later
– Slower restore speeds
– tape can be fragile if not handled with appropriate care
Pros to D2D:
– “faster recovery” of something that was very recently lost (not really a major win here since ZFS gives us snapshots so we already have this)
– Slightly cheaper cost/GB for smaller storage. Amortizes worse as you have to grow drives (replace instead of supplement)
– Can be re-purposed as a file server later (presuming you buy a box with enough power to do so)
– Potential for deduplication to reduce overall storage needs – depends on the type and quantity of data as well as the quality of dedup available (right now today in OSS I consider it still risky but I am certain it will improve later).
– quicker restore
Cons to D2D:
– Hard to do offsites (unless you want to build a duplicate system somewhere else)
– Doesn’t handle long-term archive unless you have a _huge_ file server
– must be multiple times larger than the source filesystem in order to have any longevity
– temptation to split duties on the same box (file server + backup) – very much not a good idea
– more complex
– Losing more than 3 drives in the array (concurrently) means all data lost
– can be reformatted by accident, losing all data irrecoverably
– higher power/cooling requirements