Enterprise arrays bundle premium hardware with capacity-based software licensing and costly controller upgrades. NetApp ONTAP layers per-capacity licensing and per-feature add-ons on premium hardware; Dell PowerStore charges for capacity-based software plus expensive controller and expansion upgrades; Pure Storage FlashArray locks recurring spend into its Evergreen subscription; HPE Alletra pairs capacity licensing with premium hardware and renewal-gated upgrades. In every case the software cost scales with the terabytes you hold, and the refresh cycle is tied to a support contract you do not fully control. Software-defined and open targets, TrueNAS (ZFS), Ceph (block/file/object), MinIO (object), run on commodity hardware and remove the per-TB software tax, in exchange for more design and operational ownership.
Scale versus simplicity
- TrueNAS, ZFS NAS/SAN, strong for file/iSCSI with snapshots and replication; simplest path for many estates, and the natural landing spot when you are leaving ONTAP and want a familiar file-and-block model without the licensing.
- Ceph, unified block (RBD), file (CephFS), and object (RGW); scale-out and self-healing, but a real operational commitment. It pays off at scale and for multi-protocol, growing estates, which is why it is the common destination off PowerStore, FlashArray, and Alletra when the goal is horizontal growth rather than a like-for-like appliance.
- MinIO, S3-compatible object storage for cloud-native and backup targets.
The honest split is scale versus simplicity: a small, latency-sensitive, single-protocol estate is usually better served by TrueNAS, while Ceph rewards teams that have both the scale and the appetite to run a distributed system.
Plan the data path
Inventory LUNs/volumes/shares, capacity, IOPS/throughput, snapshots, and replication/DR relationships. Confirm host connectivity (FC/iSCSI/NFS) and multipathing to the target. Provision for performance, not just capacity, match IOPS/latency, plan N+1 redundancy, and budget snapshot/replication overhead (often 20–30%).
Moving the data at the host level
There is rarely an array-to-array replication path between a proprietary source and an open target, so data moves at the host or hypervisor level. Migrate by host-level copy (rsync for files, preserving permissions/sparseness), storage vMotion for VM datastores, or application-native replication for databases. Re-present LUNs/shares from the target, update mounts/initiators, and rescan/verify multipath. On ZFS, LUNs become zvols and shares become datasets; on Ceph, block volumes become RBD images (attached directly or via the iSCSI gateway) and file shares become CephFS. For object workloads, sync with rclone/mc. Keep the source-to-target mapping one-to-one where you can, so rollback is a single re-present rather than a rebuild.
Cutting over and testing failover
Quiesce host I/O (or use online replication), run a final sync, re-present, and validate capacity, IOPS, and latency against baseline. Run a path-failure / controller-failover test and a snapshot/replication/DR test, plus data-integrity verification, before retiring the source array. Roll back by re-presenting the original LUNs if validation fails.
Start with one LUN, keep the array as fallback
Test a single LUN/share end-to-end first, then migrate host-group by host-group in maintenance windows. Keep the source array available as fallback through hypercare.
The operating-cost reality
Removing per-TB software licensing is not the same as removing cost; it shifts it from a vendor line item to commodity hardware plus your own operations. That trade is strongly favourable at scale and on long horizons, where a growing open cluster avoids repeated controller upgrades and renewal-gated refreshes. It is far less clear for small, single-protocol estates, where the minimum viable footprint of an open target, and the engineering time to run it, can erode or exceed the savings. Be honest about which side of that line you sit on, size on usable terabytes rather than raw, and count the team time to operate the new platform as a real number, not a rounding error. Ceph in particular is powerful but a genuine operational commitment; it pays off at scale, and it punishes estates that adopt it for the licensing math alone.
Use the TCO calculator to model your storage spend sized on usable TB.