Asking for help on storage tiering (Nextcloud, mostly)

wbk

Renowned Member
Oct 27, 2019
270
47
68
Hi all,

I'm trying to solve a problem revolving around my backup strategy.

I have a number of services running on a homeserver. Backupwise, the most impactful is Nextcloud. I converged on Nextcloud for all data, to be able to easily add to it (clients on different platforms, auto upload) and share either at home or to a wider audience.

What I run into:
  • Incremental backups to (remote) PBS take little space, but a lot of time
  • Incremental backups to tape take little time, but full backups take a lot of time
  • The bulk of the data never changes
    • There's no need for the PBS backup task to compare chunks of data covering 30 years of data, when only 3 months worth have slight chance of changing
    • There's no need for a full tape backup to also backup the last 30 years of data, that are not changed since the last full backup
The incremental backup runs multiple days for some 6-7 TB of data. Writing a full tape backup (LTO3 for the time being, manual loader) takes about a week for 15 tapes (not actual time, but I'm not always there the minute a tape can be changed).

My envisioned solution is to create multiple smaller storage containers. Each of them will keep 3-5 years of data, up to 1,5 TB.
  • containers for 20th century and early 21st only need to be backed up once a quarter of less often
  • containers for 2020-2025 at a higher frequency, like once a month
  • container with current year data runs every week
  • the container that runs the homeserver with Nextcloud itself is then small enough to run daily or multiple times a day
Tape backups will fit on less than 4 tapes per container, and later moving to more recent generations, only two or a single tape. Such a backup only takes a couple of hours.

I'm now migrating to a somewhat heavier server. I thought to:
  • Make 1,5 TB volumes for each container
  • Export data from the storage containers via NFS
  • Mount storage in Nextcloud on the same host
I find out that:
  • Kernel mode NFS has the host sharing data from one container
  • To a mount point in a container on the same server
  • It's not an elegant solution
Please share your thoughts on how I could better solve this. I'd like to keep using separate containers, so I can use the efficient (remote) backup and convenient tape functionality of PBS.