Self hosting 10TB in S3 on a framework laptop and disks

Original Article ↗ Hacker News Discussion ↗

What “self‑hosting” means here

  • Debate over terminology: some feel “self-hosting” should imply running network-accessible services with resiliency, not just “owning a computer.”
  • Others argue that, relative to today’s norm of cloud services, running your own S3-compatible object store absolutely qualifies.
  • Some think purely local, non-Internet-accessible services don’t quite match the usual sense of “self-hosting.”

S3 vs S3‑compatible object storage

  • Several commenters are confused by the title “self hosting 10TB in S3,” expecting Amazon’s service.
  • Clarified: this is self-hosted object storage with an S3-compatible API (Garage), not AWS S3.
  • Some find using “S3” for any compatible API misleading and prefer “S3-compatible object storage.”

Storage design, reliability, and scope

  • 10 TB is seen by some as trivial (fits on one disk; RAID1 or simple ZFS is “easy”), by others as non-trivial once resiliency, backups, and off-site redundancy are included.
  • JBOD over USB raises concerns about single points of failure and “easy to pull out” cabling.
  • ZFS is used on top of USB; discussions note ZFS is not inherently RAID and the redundancy level is unclear from the post.

Backups and acceptable risk

  • Many ask about backups; data loss is viewed as the real issue, not downtime.
  • OP reports syncing some data to cloud S3 now and planning a second physical site later.
  • Alternative strategies discussed:
    • Two independent ZFS pools with periodic snapshots and zfs send/recv instead of mirrors.
    • Mirrors plus periodically powered-on backup disks, snapshot rotation, and separate backup servers to mitigate ransomware.

Software choices: Garage, MinIO, Ceph, others

  • MinIO: multiple reports of features being removed from the free version and UI degradation; seen as pushing users to paid tiers.
  • This feeds a broader criticism of “open source cosplay” and CLAs; others push back on some license complaints (e.g., AGPL obligations).
  • Garage: some worry about low Git activity; a maintainer explains it’s stable, actively maintained for production clusters, with limited but ongoing feature work.
  • Ceph: praised for flexibility (object, block, file) but higher complexity; advice includes avoiding SMR drives and consumer SSDs.
  • Other alternatives mentioned: SeaweedFS, ZeroFS, OpenStack Swift–style systems, etc.

Hardware, noise, and appliance vs DIY

  • The Framework laptop + USB JBOD approach is seen as clever and power-efficient, but some would prefer a small server (old Dell, QNAP, NAS appliances, NUC/RPi).
  • HDD noise at this scale is noted; some recommend specific DAS enclosures or small rack/case options.
  • One camp prefers storage as an “appliance” to minimize future maintenance; others enjoy the DIY/home-lab aspect.

Filesystems and misconceptions

  • ZFS vs btrfs: some consider ZFS “RAM hungry” and fragile on USB; others reply that ZFS runs fine with modest RAM and works well even over USB, using available memory as cache.
  • Discussion around RAID levels (mirror vs raidz1 vs single-disk + snapshots) highlights the tradeoff between hardware cost, performance, and tolerance for a few hours of data loss.

Use cases for self‑hosted S3

  • Practical uses mentioned: Veeam backups, Velero/k8s backups, app logs, Android APK storage, local processing pipelines with selective syncing to cloud object storage.
  • Some argue a traditional NAS/NFS is simpler for many home needs, but others note many modern tools explicitly require an S3-like object store, making S3-compatible setups valuable.