Adding a Disk Shelf: JBODs and Enclosures
Every storage server eventually fills its last bay, and the upgrade path splits: replace drives with bigger ones, buy a second server, or attach a disk shelf — an enclosure that adds twelve or twenty-four bays to the machine you already run. Shelves are the least glamorous and most economical answer, and the refurbished market sells them for less than the caddies once cost.
What a JBOD actually is
Just a Bunch Of Disks: a chassis with power, fans, bays and a SAS expander — no controller logic, no RAID, no opinion. A cable from the server's SAS HBA to the shelf's input port, and the operating system sees the shelf's drives as if they were internal. All intelligence stays server-side, which is exactly what ZFS/TrueNAS builds want: more raw disks presented plainly. Expanders multiply connectivity — one 4-lane SAS port (∼48Gb/s on 12G SAS) fans out to dozens of drives, bandwidth shared but generous against spinning-disk realities.
The used-market cast
HPE D3610/D3710 (12LFF/24SFF, 12Gb) and their D2600/D2700 6Gb predecessors; Dell MD1200/MD1220 (6Gb) and MD1400/MD1420 (12Gb); plus the NetApp DS4246-class shelves the homelab world adopted for their bay count per dollar. Compatibility is refreshingly loose: a standard LSI/Broadcom HBA in IT mode talks to any of them, cross-brand included — the interop questions worth checking are cable connector types (SFF-8088 mini-SAS vs SFF-8644 HD) and, on some shelves, drive interposer/firmware quirks documented by their communities.
Cabling and redundancy choices
Single path is simple: HBA → shelf IN, done; daisy-chain the next shelf from OUT. Dual-expander shelves offer two paths per drive — wire both to (ideally) two HBAs and multipath-capable software rides through a cable or expander failure. For a backup pool, single-path is honest; for a primary pool you would rather not take down for a cable, dual-path costs one more HBA and cable. Either way: shelves power on before the server, labels go on both cable ends, and the chain stays short enough that one shelf's bandwidth share still exceeds its drives' appetite.
Growth patterns that stay sane
With ZFS, add capacity in whole vdev units per shelf (a shelf holding two six-wide RAIDZ2 groups mirrors the classic internal layout) rather than scattering mixed groups across enclosures — failure domains stay legible. Keep the boot and hot tiers inside the server; let shelves carry the bulk tier. And before buying, run the arithmetic that makes shelves shine: an R730xd's twelve internal bays plus one D3610 equals twenty-four LFF drives behind one machine — a quarter-petabyte raw with commodity 12TB disks, for the price of one new workstation.