Hard disks are usually accessed using a block device node (sometimes called a cooked device node). When accessing a device as a block, data transmitted to or from the device is buffered, or collected until there is sufficient data to make accessing the device worth the trouble. Block devices are generally considered more efficient than raw devices.

The device nodes for block devices are named after the device driver; for example, /dev/wd3.

Raw devices are sometimes called character devices, because they access a device one character at a time. If you need to control exactly how data appears on a disk (for example, when creating a filesystem) use a raw device. Raw device nodes have an r in front of their name, as in /dev/rwd3.

Raw devices do no buffering. When you tell your system to write to a raw device, the data is transmitted immediately. Raw mode works best with software that provides its own buffering or that wants to arrange data in a specific way.

Here’s an easy way to remember the difference between block and raw throughput: Say you spill a bottle of aspirin. If you pick up each aspirin individually and deposit it directly in the bottle, you’re doing an unbuffered, or raw, transfer. If you pick up the aspirin with your right hand and collect them in your left, then dump a bunch into the bottle at once (along with all of the dirt from your floor), you’re doing a buffered transfer.^[17]

Address disks (and many other devices) as raw or block by choosing the corresponding device node. Some programs expect to access raw devices, while others expect block devices. If a program opens /dev/sd1a, it’s accessing partition a on disk sd1 as a block device. If it opens /dev/rsd1a, it’s accessing the exact same partition as a character device.

Regardless of the mode, the underlying hardware remains the same; the only thing that changes is how you exchange information with the device.