HBAs — Fibre Channel, SAS & NVMe-oF
Complete buyer guide & protocol reference

A Host Bus Adapter (HBA) presents storage to the server without any processing — it is a "dumb" pass-through that connects the server to the storage device or fabric. The OS or storage software manages the drives directly. A RAID controller sits between the server and drives and performs RAID operations in dedicated hardware, presenting a logical volume to the OS. For Fibre Channel SAN connections, always use an HBA — you never want RAID between the server and a storage array that manages its own RAID. For local SAS/SATA drive expansion, HBA in "IT mode" is required for ZFS/Ceph; RAID mode is used for simpler local storage.

Fibre Channel is a dedicated high-speed network protocol designed exclusively for storage block I/O, operating over optical fibre (and occasionally copper). FC provides inherently lossless, in-order delivery without the tuning complexity of lossless Ethernet for RDMA. FC SANs are highly mature — most enterprise storage arrays (Pure Storage, NetApp, HPE, Dell, Hitachi) support FC connectivity. FC generations: Gen 3 = 8Gb, Gen 5 = 16Gb, Gen 6 = 32Gb, Gen 7 = 64Gb. The FC fabric consists of the HBAs (in servers), FC switches (Brocade/Cisco MDS), and storage array FC ports.

Traditional Fibre Channel carried SCSI commands (FC-SCSI / FCP). FC-NVMe replaces SCSI with the NVMe command set over the same FC fabric — same switches, same optics, just a firmware/driver update on the HBAs and storage arrays. FC-NVMe achieves significantly lower latency than FC-SCSI because NVMe's command queue architecture is far more efficient than SCSI. Pure Storage FlashArray, NetApp AFF, and Dell PowerStore all support FC-NVMe. Existing 16Gb+ FC fabric infrastructure can support FC-NVMe without replacing switches or optics.

NVMe-oF extends NVMe over an IP network using RDMA (RoCEv2) or TCP as the transport, as an alternative to FC. NVMe/RDMA (RoCEv2) achieves latency comparable to FC-NVMe (10–20µs) but requires a lossless Ethernet fabric. NVMe/TCP runs over standard Ethernet without the lossless requirement — simpler to deploy but adds ~50µs of latency overhead from TCP processing. For new deployments with modern Ethernet infrastructure, NVMe/RDMA over 100GbE is increasingly chosen over FC due to lower total cost of ownership (no dedicated FC switches).

SAS HBAs can operate in two modes: IR (Integrated RAID) mode, where the card performs RAID, or IT (Initiator Target) mode, where the card is a pure passthrough and presents each drive directly to the host OS. For ZFS (TrueNAS, OpenZFS), Ceph, and software RAID, IT mode is mandatory — ZFS requires direct drive access to its own error correction and write path management. A RAID controller in IT mode is sometimes called a "flashed to IT mode" card. Broadcom 9405W, 9400-16i, and similar cards are common in storage server builds for exactly this purpose.

FC zoning is the Fibre Channel equivalent of VLANs — it defines which HBAs (initiators) can see which storage ports (targets) in the fabric. Without zoning, every server in the SAN can see every storage port, creating a security risk and potential for incorrect LUN masking. Hard zoning (by port number) is the standard — each server's HBA is zoned to specific storage array ports. This is configured on FC switches (Brocade or Cisco MDS), not on the HBAs themselves. Zoning is a mandatory step in any new FC SAN deployment.