Here's a question regarding iSCSI. What if my target drives are Serial ATA or a non-SCSI interface? Do I have to translate my SCSI commands at the target end? SCSI drives do not connect directly to Ethernet fabric like they do on a Fibre Channel SAN, so why run SCSI over Ethernet? Doesn't this require a specialized NIC or driver to strip the SCSI commands out of the TCP/IP stream?
There a number of topics here. First, don't expect non-SCSI drives to work with i-SCSI, unless you get a SCSI-to-ATA interface solution such as those developed by 3-ware. Other subsystem controllers are likely to emerge, too. The "SCSI" part refers to standard SCSI syntax.
Second, for the most part, devices are not expected to have Ethernet interfaces soon, although a lot of R&D work has been done in this area at Carnegie Mellon, UC Berkeley, and other places. Maybe I should say don't expect i-SCSI disk drives with native Ethernet interfaces in abundance soon.
For the most part, iSCSI is an upper-layer network protocol technology that will need to interface with storage subsystems. Although it should not require
specialized NICs, for practical purposes, the NIC should have TCP and i-SCSI
acceleration in hardware (none exist yet) to offload the heavy TCP protocol
processing anticipated for storage traffic. After the transfer is processed by
the subsystem controller at the target side, almost any kind of drive could be
used inside the subsystem including ATA, SCSI, or 1394 (whatever).
This was first published in July 2001