In “Design Rules” [5], Baldwin and Clark apply a real options theory to study modularity in the computer industry. They show how modularity in computer systems (like the IBM 360) changed the industry tremendously.
Modularly designed computers consist of components that have defined interfaces. Because each component conforms to its interface rules, mod- ules that follow the defined interface are interchangeable. In contrast, an interconnected system has no swappable components because only a sin- gle massive component exists. Baldwin’s work shows how modularity increases value and how increasing technology uncertainty about the value of the modules increases this value of modularity.
To see how a modular design provides value, consider the evolution of a computer system without a module design. Figure 5.3 illustrates such a system: It consists of four main functions — storage, memory, I/O, and the CPU. Suppose that this computer is being redesigned and both the mem- ory and the CPU are changed. Now, assume that this redesigned CPU worked well and increased the value of the total system by +1; however, the new memory design did not work as expected. It decreased the value of the total system by –2. When redesigning a computer that has its func- tional pieces interconnected, the new artifact provides a single choice; the new system performs, as a whole, better, worse, or the same than its pre- decessor does, and you can take it or leave it. In this case, the new system has a value less than the original system; it is the failed memory experi- ment that drags down the total value of the system. The interconnected architecture of this computer does not allow the choice of using only the improved CPU, without the inferior new memory design.
Figure 5.3 Interconnected system.
Value = 0
CPU Memory
I/O Storage
Value = 0
CPU Memory
I/O Storage
Value = -1 Value of experiments
OR
CPU Memory
I/O Storage Memory
CPU
= -2
= +1
Intuitive View of Options Theory 79
In contrast to the interconnected design previously mentioned is the mod- ular computer system illustrated in Figure 5.4. If the same redesign is attempted with this modular architecture, there are more choices for the new system when compared to the interconnected system, a similar effect to hav- ing more possible outcomes with staged development, as discussed previ- ously. The system with the most value uses only the new CPU, keeping the older, but better-performing memory design. This new system has a value of +1, higher than the interconnected system allows. As with all options-like sit- uations, increased uncertainty implies increased value of the modulariza- tion. The modularity allows the system designer to pick and choose the components of the new system, thus maximizing the value. Uncertainty increases this value because as it grows it increases the potential of a really great choice emerging. The modular design increases value by providing a portfolio of options rather than the less valuable option on a portfolio.
Baldwin and Clark [5] computed the value of modularity. Let V1be the value of a complex system built as a single module, and let Vjbe the value of the same system with jmodules. If the cost of modularity is ignored, then the value of dividing a complex system into j components is: Vj = j1/2V1. That is, the modularized system exceeds the value of the intercon- nected design by the square root of the number of modules. This value does not depend on the variance of the distribution because for each mod- ule there is only one choice — keep the old or use the new — and there is only one choice for the new module. This is intuitive; if you take a single sample from a random distribution, the expected value is not dependent on the variance. If there are many choices for the new module, then the variance of the distribution is important to the expected value of the best of many choices.
This modularization allows the designers to experiment on modules that have the most potential for altering the value of the system. Each experiment is one design of the module. Performing many experiments on the compo- nents most critical to overall system performance maximizes the overall value. Because of the modular design, the designer now has the option of picking the best outcome from many trials. For example, suppose the designers of a new computer system need to increase the rate at which a CPU module processes instructions, as illustrated in Figure 5.5. (This is sim- ilar to the idea behind Figure 6.1 in the next chapter.) It shows that three attempts are made to improve the CPU; the worst experiment lowers the value of the total system by –2, and the best new design increases the total 80 Chapter 5
system value by +2. By attempting several technically risky new technolo- gies for a CPU, the designer can improve the odds of reaching the goal of faster instruction execution. In this case, the best CPU increases the value of the total system by +2. The modularity allows system designers to focus on components that have the greatest potential to increase the value of the whole system.
Figure 5.4 Value of modular system.
CPU Memory
I/O Storage Value = 0
CPU Memory
I/O Storage Value = -2 OR
OR Value = +1
OR Value = -1
CPU Memory
I/O Storage Value = 0
Value of experiments Memory
CPU
= -2
= +1
CPU Memory
I/O Storage
CPU Memory
I/O Storage
Intuitive View of Options Theory 81
Figure 5.5 Value of experimenting where it counts most.
This modularity allows picking the best from many experiments on a module, which is illustrated in Figure 5.5. In the preceding example there were three experiments for the CPU, and the best of these experiments yielded +2. Assuming that the experiments on the CPU are normally dis- tributed, then the values of the experiments are expected to look like Fig- ures 6.1 and 6.2 in the next chapter. Each experiment creates a choice to use a particular module from many different choices, which is the same as picking the best of many experiments. As the uncertainty about the tech- nology used in the module grows, so does the value of having a choice from many modules. If uncertainty is high, and if you have enough exper- iments, it becomes likely that an exceptional (as defined as standard devi- ations of +3 or more from the mean) new module is found.
This value of picking the best module from many choices is similar to the value gained from a distributed management structure because the exper- imentation it allows enables users to have many choices of network-based services. Interconnected systems make experimentation difficult much the same way that central management hampers the ability to experiment.
Modularity gives designers choices in the same way that distributed man- agement gives users choices. This model of the value of modularity forms
CPU(2) Memory
I/O Storage Value = +2
CPU Memory
I/O Storage Value = 0
CPU(1) Value = -2
Value = 2
Value = +1 CPU(2)
CPU(3) 82 Chapter 5
the framework for this book because of this mapping from the value of modularity to the value of distributed management. Both have the most value in high uncertainty because of the increased value of experimenta- tion enabled by module design and distributed management structure.