The Impact of Component Modularity on Design Evolution: Evidence from the Software Industry

The Impact of Component Modularity on Design Evolution: Evidence from the Software Industry by Alan MacCormack, John Rusnak and Carliss Y. Baldwin

Much academic work asserts a relationship between the design of a complex system and the manner in which this system evolves over time. In particular, designs which are modular in nature are argued to be more “evolvable,” in that these designs facilitate making future adaptations, the nature of which do not have to be specified in advance. In essence, modularity creates “option value” with respect to new and improved designs, which is particularly important when a system must meet uncertain future demands.

Despite the conceptual appeal of this research, empirical work exploring the relationship between modularity and evolution has had limited success. Three major challenges persist: first, it is difficult to measure modularity in a robust and repeatable fashion; second, modularity is a property of individual components, not systems as a whole, hence we must examine these dynamics at the microstructure level; and third, evolution is a temporalphenomenon, in that the conditions at time t affect the nature of the design at time t+1, hence exploring this phenomenon requires longitudinal data.

In this paper, we tackle these challenges by analyzing the evolution of a successful commercial software product over its entire lifetime, comprising six major “releases.” In particular, we develop measures of modularity at the component level, and use these to predict patterns of evolution between successive versions of the design. We find that modularity has a strong and unambiguous impact on design evolution. Specifically, we show that i) tightly-coupled components are “harder to kill,” in that they have a greater likelihood of survival in subsequent versions of a design; ii) tightly-coupled components are “harder to maintain,” in that they experience more surprise changes to their dependency relationships that are not associated with new functionality; and iii) tightly-coupled components are “harder to augment,” in that the mix of new components added in each version is significantly more modular than the legacy design.

Read Paper

Evolution Analysis of Large-Scale Software Systems Using Design Structures Matrices and Design Rule Theory

Evolution Analysis of Large-Scale Software Systems Using Design Structure Matrices and Design Rule Theory by Matthew J. LaMantia, Yuanfang Cai, Alan David MacCormack and John Rusnak

Designers often seek modular architectures to better accommodate expected changes and to enable parallel development. However, we lack a formal theory and model of modularity and software evolution, which can be used for description, prediction, and prescription. According to Baldwin and Clark’s  theory, modular architectures add value to system designs by creating options to improvethe system by substituting or experimenting on individual modules. In this paper, we evaluate their theory by looking at the design evolution of two software product platforms through the modeling lens of design structure matrices (DSMs) and design rule theory. Our analysis shows that DSM models and options theory can explain how real-world modularization activities in one case allowed for different rates of evolution in different software modules and in another case conferred distinct strategic advantages on a firm (by permitting substitution of an at-risk software module without substantial change to the rest of the system). The experiment supports our hypothesis that these formal models and theory can account for important aspects of software design evolution in large-scale systems.

Read Paper

Hidden Structure: Using Network Methods to Map System Architecture

“Hidden Structure: Using Network Methods to Map System Architecture” by Carliss BaldwinAlan MacCormack and John Rusnak

In this paper, we describe an operational methodology for characterising the architecture of complex technical systems and demonstrate its application to a large sample of software releases.  Our methodology is based upon directed network graphs, which allows us to identify all of the direct and indirect linkages between the components in a system. We use this approach to define three fundamental architectural patterns, which we label core-periphery, multi-core, and hierarchical. Applying our methodology to a sample of 1,286 software releases from 17 applications, we find that the majority of releases possess a “core-periphery” structure. This architecture is characterized by a single dominant cyclic group of components (the “Core”) that is large relative to the system as a whole as well as to other cyclic groups in the system. We show that the size of the Core varies widely, even for systems that perform the same function. These differences appear to be associated with different models of development—open, distributed organizations develop systems with smaller Cores, while closed, co-located organizations develop systems with larger Cores. Our findings establish some “stylized facts” about the fine-grained structure of large, real-world technical systems, serving as a point of departure for future empirical work.

Read Paper