Categories: PCs and Hardware

Continued from SCSI Flavors

The SCSI specification defines both the details of the SCSI bus and the protocol by which data transfer takes place between devices. A SCSI device has a built-in controller that interprets commands from the host adapter, much like an IDE or EIDE drive, which interacts with the IDE adapter card. But SCSI adapters are more complex than IDE adapters, because the latter rely on the CPU for functions such as data transfer between devices. The SCSI adapter, on the other hand, knows all the possible SCSI commands and doesn't depend on the CPU for any of them. That's why it's a good idea to upgrade your SCSI adapter if you buy a new SCSI device with a built-in controller that supports the latest SCSI version. For the most part, devices are backward-compatible, but their efficiency and feature sets may well rely on the ability to interpret the newest SCSI commands, which your old host adapter might not understand.

SCSI was designed as a parallel interface that allowed as many as eight devices to be connected along a single cable. The cable and the host adapter form the SCSI bus, which operates independently of the rest of the computer. The bus allows data exchange among devices without taking up CPU cycles and thus without relying on the system bus. That makes the SCSI bus's potential speed higher than that of the bus for interfaces such as IDE, which at least partly rely on the system bus. Tape backups, for example, from a SCSI hard disk to a SCSI tape drive can perform extremely well in the background, if the backup software is designed for full SCSI support. Anything that demands CPU and system bus activity, however, requires interaction between the SCSI host adapter and the system.

Through the SCSI BIOS, each device is given a distinct address or SCSI ID, ranging from 0 to 7 for a narrow (8-bit) bus or 0 to 15 for a wide (16-bit) bus. Devices that request I/O processes are called initiators. Devices that perform operations requested by initiators are called targets. Through its built-in controller, each target can accommodate up to eight other devices; these are known as logical units, and they are each assigned a logical unit number, or LUN. Commands to the SCSI controller identify devices according to their LUNs; the controller connects to the SCSI bus and executes the commands.

Each SCSI address is a target, and since each of these targets is also a SCSI device with its own controller, it can assign addresses to additional SCSI devices. In most simple systems, the host adapter card--which is the first target in the system--acts as a controller for itself and 7 (or 15) additional SCSI addresses. Let's say it's controlling a hard disk, a CD-ROM drive, and a scanner. The hard disk would be assigned SCSI ID 0, the CD-ROM drive SCSI ID 1, and the scanner SCSI ID 2. The LUN for each of these addresses would be 0, since the devices (the logical units) are all closed, single-function hardware devices. So SCSI ID 0 controls the hard disk and assigns it LUN 0, since the disk is the first and only logical unit on that particular address. SCSI ID 1--a completely separate address--assigns LUN 0 to the CD-ROM drive, because the CD-ROM drive is the first and only logical unit on that address. And so on. Each of these SCSI IDs is capable of controlling more devices (LUN 1 through LUN 7). With closed, single-function devices such as disk drives, however, this capability lies completely unused. One of the goals of SCSI developers and manufacturers is to create products that can make use of this wasted capacity, primarily through a bus bridge, a two-port device that connects to the main SCSI bus and creates an auxiliary bus to which additional logical units can be attached.

Next: How SCSI Works

Published as Tutor in the 3/10/98 issue of PC Magazine.