Computer Organization

CSC 105 - The Digital Age - Weinman

Summary:: We discuss the basics of the computer's central processing unit (CPU) architecture and overall computer organization.

We have examined computation at many different levels this semester. At the lowest level, we know how computers store and represent data in binary bits. We also have seen how primitive digital circuits can implement simple computations and storage (e.g., addition and flip-flop memory). At a much higher level, you've written algorithms for solving a variety of problems. In this reading and the next, we'll cover the intermediate stages to learn how the smaller, low-level pieces made up of bits and circuits combine to make larger components that can be programmed.

Organization

There are essentially four major components (or categories) that comprise a computer:

CPU (Central Processing Unit): This piece is what actually executes programs and performs the central computations.
RAM (Random Access Memory): This piece stores data, like the list of numbers being used in a binary search or the bytes of an ASCII encoded text file. It also stores the written instructions for a program or algorithm.
Peripherals: These are just about any other major components you might want to connect to the CPU or RAM to get input from for computing with or send a computational result to as output. Examples might include the video screen, a keyboard or mouse, hard disks, or a network port.
Buses: These are the wires that connect and transport data among the aforementioned components.

Memory Hierarchy

Once we begin to talk about actually building a computer, it turns out that how you store data is nearly as important as how you represent data. There are a variety of ways to implement data storage on a computer. Just as with information stored in the "real world," the spectrum of performance ranges from very fast, expensive, low-capacity components to much slower, cheaper, and larger capacity components. This spectrum represents the memory hierarchy.

At the highest level, data stored in a memory that uses logic gates like those we have studied are called registers. Much like your brain, data in registers may be accessed very quickly. Registers are very expensive, thus computers tend to have relatively few of them. Because they are so fast, they tend to be closest to where the computation happens: in the CPU.

At the next level is cache. This form of memory is a little bit slower than registers, but data in the cache is still relatively quick to access, much like a sheet of paper in front of you on the desk. Cache accesses might take anywhere from five to twenty times as long as a register access.

Below the cache in the hierarchy is the RAM. This is slower yet and farther from the CPU, yet it is where most of the computer's data is stored. This might be more like reading something from a book in your office. Accessing data stored in RAM may take anywhere from twice to twenty-four times as long as a cache access. In turn, the CPU could wait over one-hundred times as long for data stored in RAM as it might for data stored in its registers. But there is still another level of the hierarchy.

The RAM available to most computers today has grown substantially with the concomitant decrease in price. However, as we operate on larger and larger amounts of data, it seems never to be enough. Therefore, when a computer program requires more data than will fit in its available RAM, it will often resort to the hard disk, typically the last stage of a memory hierarchy. The capacities of hard drives are often one-thousand times bigger than the amount of available RAM. However, this capacity comes at the cost of being over twenty-million times slower than a register. Such a delay is worse than getting a book from the library; it's more like inter-library loan.

The following table summarizes the properties of the memory hierarchy. In the next section we'll discuss further details of the registers and RAM, ignoring cache for now and deferring our discussion of disks for another week.

Hierarchy	Speed	Capacity	Cost ($/bit)	Distance
Registers	Fastest	Smallest	Highest	Nearest CPU
Cache
RAM
Disk	Slowest	Largest	Cheapest	Farthest from CPU

CPU

The central processing unit is where most of the computational "action" happens, so to speak. Here, we'll look at the major components of the CPU and what their duties are. These are pictured in the figure below.

First up is the control unit. This piece of the CPU coordinates all of its activities, controlling program execution. In essence, it translates instructions (usually written by a programmer) into physical signals that cause them to be executed. This typically means sending appropriate signals through the ALU (arithmetic/logic unit), which performs the most basic operations on data, such as addition or computing a bit-wise logical AND between two binary words.

Of course, the data being added or ANDed must come from somewhere, hence the need for general-purpose registers. As we mentioned, these form the highest level of the memory and are actually within the CPU. These registers (denoted R₁,R₂,...,R_N in the figure) store the operands (inputs) used by the ALU. Similarly, the ALU can send it results back to store them in these registers.

In addition to the general-purpose registers, there are two special-purpose registers in the CPU, namely the IR (instruction register) and the PC (program counter). The IR stores the instruction currently being executed. This is then read by the control unit, which configures the ALU (say to subtract, rather than add) and the buses connecting all of the pieces so that the intended result is computed on the correct input registers and stored in the correct output register. The PC stores the address in memory where the next instruction to be executed is located. Speaking of a memory address, what is that? We discuss that next.

RAM

RAM is a large set of memory cells. Typically each cell consists of multiple bytes, called a word. In modern computers, words are often made of four bytes, but this can vary substantially, depending on whether the computer is a handheld tablet or a top-of-the-line data crunching server.

Each of these memory cells has an "address" or a location. Why is this called "random access"? It is because we can access the data at any location at any time simply by specifying its address. Contrast this with something called "sequential access." As you might guess, sequentially accessible data required you to march from the start of the data through to the address you wanted. Think about the difference between fast-forwarding on a cassette tape or VHS (sequential formats) to find what you want and simply requesting a certain track on a CD or DVD (random access formats). One may visualize RAM as a table of data, each line numbered with the address of that data, as shown below. A program can then dynamically read or write the contents of any RAM cell.

Address	Data Value

0	42
1	255
2	0
...

You may have noticed that the CPU as we have described it can only operate directly on data that resides in its general-purpose registers. However, there are not very many of these registers. Therefore, we will need a way to access the much larger amount of storage available in RAM.

Memory Bus

Finally, we discuss how data is transferred between the CPU's registers and RAM (typically referred to simply as memory). The memory bus is a collection of wires that facilitate this communication. Reading from memory typically consists of the following three steps.

The CPU sends an address over the bus to the memory component.
The memory responds to this request by sending back the data at that address to the CPU.
Finally, the CPU stores this data in one of its registers.

As we noted earlier, the ALU may have computed hundreds of operations using data in its local registers by the time data arrives from memory. Thus, one important job of early programmers (and now compiler programs) was managing the data to keep the most immediately needed values in the registers, rather than in RAM.

Writing data to memory operates similarly, except the CPU sends both the address and the data across the bus to the RAM, which then stores the data at the appropriate location. Depending on the architecture of the computer, the CPU may not necessarily need to wait for this operation to complete before proceeding.

On the accompanying lab, you will experiment by manually configuring registers and the ALU (playing the control unit), as well as reading and writing data in memory in a small computer simulator.

Acknowledgments

Adapted from materials by Marge Coahran. Used by permission.

This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 United States License.