Characteristic of Typical I/O Devices

Consider, for example, the keyboard as an input device. The keyboard is basically a character-based device. You are probably already aware that typing on the keyboard of your PC results in Unicode or ASCII input to the computer, one character at a time. Even mainframe terminals, many of which can send text to the computer a page at a time, only transmit a page occasionally, so the data rate for keyboards is obviously very slow compared to the speed at which the CPU processes the data.

Input from the keyboard is very slow because it is dependent on the speed of typing, as well as on the thought process of the user. There are usually long thinking pauses between bursts of input, but even during those bursts, the actual input requirements to the computer are very slow compared to the capability of the computer to execute input instructions. Thus, we must assume that if the computer is simply performing a single task, it will spend most of its time waiting for input from the keyboard.

It is also useful to note that there are two different types of keyboard input. There is input that is expected by the application program in response to a ‘‘read’’ statement of some kind requesting input data for the program. Then there are other times when the user wishes to interrupt what the computer is doing. On many computers, a character such as Control-‘‘C’’ or Control-‘‘D’’ or Control-‘‘Q’’ can be typed to stop the program that is running.

Another input device that will generate unexpected input is the mouse. When you move the mouse, you expect the cursor to move on the screen. Clicking on a mouse button may serve as expected input to a program, or it may be unexpected and change the way in which the program is executing.

Printers and display screens must operate over a wide range of data rates. Although most monitors and printers are capable of handling pure ASCII or Unicode text, most modern output is produced graphically or as a mixture of font descriptors, text, bitmap graphics, and object graphics, a page or a screen at a time, using a page description language.

Contrast the I/O requirements of keyboards, screens, and printers with those of disks and DVDs. Since the disk is used to store programs and data, it would be very rare that a program would require a single word of data or program from the disk. Disks are used to load entire programs or store files of data. Thus, disk data is always transferred in blocks, never as individual bytes or words. Disks may operate at transfer rates of tens or, even, hundreds of megabytes per second. As storage devices, disks must be capable of both input and output, although not simultaneously. On a large system there may be several disks attempting to transfer blocks of data to or from the CPU simultaneously. A DVD attempting to present a full screen video at movie rates without dropouts must provide data steadily at input rates approaching 10 megabytes per second, with some transient rates and high definition video rates even higher. In addition, video and audio devices require a steady stream of data over long periods of time. Contrast this requirement with the occasional bursts of data that are characteristic of most I/O devices.

For both disk and image I/O, therefore, the computer must be capable of transferring massive amounts of data very quickly between the CPU and the disk(s) or image devices. Clearly, executing a single instruction for each byte of data is unacceptable for disk and image I/O, and a different approach must be used. Furthermore, you can see the importance of providing a method to allow utilization of the CPU for other tasks while these large I/O operations are taking place.

With the rapid proliferation of networks in recent years, network interfaces have also become an important source of I/O. From the perspective of a computer, the network is just another I/O device. In many cases, the network is used as a substitute for a disk, with the data and programs stored at a remote computer and served to the local station. For the computer that is acting as a server, there may be a massive demand for I/O services.

It should be pointed out that disks, printers, screens, and most other I/O devices operate almost completely under CPU program control. Printers and screens, of course, are strictly output devices, and the output produced can be determined only by the program being executed. Although disks act as both input and output devices, the situation is similar. It is the executing program that must always determine what file is to be read on input, or where to store output. Therefore, it is always a program executing in the CPU that initiates I/O data transfer, even if the CPU is allowed to perform other tasks while waiting for the particular I/O operation to be completed.

A table of typical data rates for various I/O devices appears in Figure 9.1. The values given are rough approximations, since the actual rates are dependent on the particular hard- ware systems, software, and application. As computer technology advances, the high end data rates continue to increase at a rapid pace

The discussion in this section establishes several requirements that will have to be met for a computer system to handle I/O in a sufficient and effective manner:

• There must be a means for individually addressing different peripheral devices.
• There must be a way in which peripheral devices can initiate communication with the CPU. This facility will be required to allow the CPU to respond to unexpected inputs from peripherals such as keyboards, mice, and networks, and so that peripherals such as printers and floppy disk drives can convey emergency status information to the executing program.
• Programmed I/O is suitable only for slow devices and individual word transfers. For faster devices with block transfers, there must be a more efficient means of transferring the data between I/O and memory. Memory is a suitable medium for direct block transfers, since the data has to be in memory for a program to access it. Preferably this could be done without involving the CPU, since this would free the CPU to work on other tasks.
• The buses that interconnect high-speed I/O devices with the computer must be capable of the high data transfer rates characteristic of modern systems. 
• Finally, there must be a means for handling devices with extremely different control requirements. It would be desirable if I/O for each of these devices could be handled in a simple and similar way by programs in the CPU.

The last requirement suggests that it is not practical to connect the I/O devices directly to the CPU without some sort of interface module unique to each device.

• The formats required by different devices will be different. Some devices require a single piece of data, and then must wait before another piece of data can be accepted. Others expect a block of data. Some devices expect 8 bits of data at a time; others require 16, 32, or 64. Some devices expect the data to be provided sequentially, on a single data line. Other devices expect a parallel interface. These inconsistencies mean that the system would require substantially different interface hardware and software for each device.
• The incompatibilities in speed between the various devices and the CPU will make synchronization difficult, especially if there are multiple devices attempting to do I/O at the same time. It may be necessary to buffer the data (i.e., hold it and release part of it at particular times) to use it. A buffer works something like a water reservoir or tower. Water enters the reservoir or tower as it becomes available. It is stored and released as it can be used. A computer buffer uses registers or memory in the same way.
• Although the I/O requirements for most devices occur in bursts, some multimedia, video and audio in particular, provide a steady stream of data that must be transferred on a regular basis to prevent dropouts that can upset a user. I/O devices and the interconnections that support multimedia services must be capable of guaranteeing steady performance. This often includes network interfaces and high-speed communication devices as well as such devices as cameras, since networks are frequently used to supply audio and video. (Think of downloading streaming video from the Web.)
• Devices such as disk drives have electromechanical control requirements that must be met, and it would tie up too much time to use the CPU to provide that control. For example, the head motors in a disk drive have to be moved to the correct disk track to retrieve data and something must continually maintain the current head position on the track once the track is reached. There must be a motor controller to move the print heads in an inkjet printer across the paper to the correct position to print a character. And so on. Of course, the requirements for each device are different

The different requirements for each I/O device plus the necessity for providing devices with addressing, synchronization, status, and external control capabilities suggest that it is necessary to provide each device with its own special interface. Thus, in general, I/O devices will be connected to the CPU through an I/O module of some sort. The I/O module will contain the specialized hardware circuits necessary to meet all the I/O requirements that we established, including block transfer capability with appropriate buffering and a standardized, simple interface to the CPU. At the other interface, the I/O module will have the capability to control the specific device or devices for which it is designed.

The simplest arrangement is shown in Figure 9.2. I/O modules may be very simple and control a single device, or they may be complex, with substantial built-in intelligence, and may control many devices. A slightly more complex arrangement is shown in Figure 9.3.I/O modules that control a single type of device are often called device controllers.