← Atlas
Atlas expedition 4 angles ↓

Replacing a Synology NAS in 2026: hardware, OS, migration, and remote access

End-to-end framework to leave a Synology NAS in 2026: a hardware shortlist by profile, OS/storage picks scored on lock-in, a DSM migration playbook, and a default remote-access stack — converging on one concrete recommended build.

4 succeeded 246 sources ~38 min read

The trigger across all four angles is the same: Synology’s 2025-Plus drive lock-in pulled ServeTheHome’s recommendation [1], and even though DSM 7.3 reversed the 3.5″ HDD restriction in October 2025, M.2 SSDs remain HCL-gated and the trajectory is what convinced operators to leave [2] [3].

Hardware and OS pull in opposite directions on lock-in. The hardware survey rewards prebuilt boxes — UGREEN DXP4800 Plus at $620 with 10GbE and BYO-OS support [4], Aoostar WTR Max at $649-699 with 6 SATA + 5 NVMe + dual 10GbE SFP+ [5], Beelink ME mini at $209-329 idling at 7-10W [6]. The OS survey scores TrueNAS Community Edition lowest on lock-in only at the app layer: the 24.10 k3s→Docker Compose rewrite with a hard 1 Jun 2025 auto-migration deadline [7] [8] is the clearest reminder that any appliance OS — not just DSM — eventually rewrites its app stack. Unraid wins the volume layer (every array disk is a standalone XFS/btrfs/ZFS filesystem, mountable on any Linux box [9]) but charges a paid licence; plain Debian + ZFS + Cockpit + zrepl wins everywhere except hand-holding [10] [11].

The migration playbook constrains hardware sizing. All three operational children agree the cutover is two-machine, side-by-side [12] — you cannot reuse SHR pools in-place on TrueNAS or Unraid because mdraid+LVM+btrfs disks must be wiped and re-pooled [13]. That implies enough disk capacity on day one for the full dataset, and rules out incremental drive-swap strategies. Hyper Backup repositories will not migrate — they are a proprietary deduplicated .HBK format readable only by DSM [14], so move live data over rsync-in-tmux at ~140 MB/s [15], not the backup archives.

ECC is a red herring at home. OpenZFS docs recommend “8GB+”, not the folkloric 1 GB-per-TB rule [16], and Matt Ahrens’s well-known position is that ZFS is no more ECC-dependent than any other checksumming filesystem [17]. If your platform supports ECC (ASRock Rack B650D4U at ~$264 with onboard BMC [18]), use it; if it doesn’t, ZFS without ECC still beats ext4 without ECC.

Concrete recommended stack — the centre of all four children:

  • Hardware: UGREEN DXP4800 Plus (turnkey, 4 SATA + 2 M.2, 10GbE, ~24W idle [19], 2-year warranty [20]) — or DIY Jonsbo N3 + CWWK i3-N305 + 32GB DDR5 at $650-770 diskless [21] if you want full BYO from day one.
  • OS + storage: TrueNAS Community Edition on ZFS, with raw zfs send -w replication chosen from day one (a single non-raw receive permanently blocks future raw incrementals [22]). Pick Unraid 7.x instead if mixed-size disks and a 2000+ container app store outweigh ZFS’s checksum guarantees [23].
  • Apps: Immich for Photos [24], Seafile for Drive [25], Frigate for Surveillance Station [26], Proxmox Backup Server for Active Backup for Business [27].
  • Remote access: Tailscale as subnet router on the NAS (replaces QuickConnect, traverses CGNAT [28]) + Cloudflare Tunnel for public services + NUT + Uptime Kuma + a smart plug — recurring cost ~$10/yr domain plus electricity [29].

Open questions left after all four angles: how much of the DSM ergonomics tax (1h/month TrueNAS maintenance vs. near-zero on DSM) the migrating operator is actually willing to absorb — and whether DSM 8 (still slipping past Q4 2025 [30]) ships anything that changes the calculus before the new hardware is bought.

Sub-topics