The Dark Side of Modularity: How Decomposing Problems can Increase System Complexity

Abstract

Decomposition is a dominant design strategy because it enables complex problems to be broken up into loosely-coupled modules that are easier to manage and can be designed in parallel. However, contrary to widely held expectations, we show that complexity can increase substantially when natural system modules are fully decoupled from one another to support parallel design. Drawing on detailed empirical evidence from a NASA space robotics field experiment we explain how new information is introduced into the design space through three complexity addition mechanisms of the decomposition process: interface creation, functional allocation, and second order effects. These findings have important implications for how modules are selected early in the design process and how future decomposition approaches should be developed. Although it is well known that complex systems are rarely fully decomposable and that the decoupling process necessitates additional design work, the literature is predominantly focused on reordering, clustering, and/or grouping based approaches to define module boundaries within a fixed system representation. Consequently, these approaches are unable to account for the (often significant) new information that is added to the design space through the decomposition process. We contend that the observed mechanisms of complexity growth need to be better accounted for during the module selection process in order to avoid unexpected downstream costs. With this work we lay a foundation for valuing these complexity-induced impacts to performance, schedule and cost, earlier in the decomposition process.

The impact of system representation choices on architecting insights

“…There are a wide variety of system architecture properties that could be measured including a wide variety of other graph theoretic measures 10 . The measures that we chose are commonly used within the field and are the subject of numerous papers 23,40,41,52–54 . We chose two contrasting measures because their focus on different system features means that any common insights across them increase the generalizability our findings.…”

“…10 The measures that we chose are commonly used within the field and are the subject of numerous papers. 23,40,41,[52][53][54] We chose two contrasting measures because their focus on different system features means that any common insights across them increase the generalizability our findings. Furthermore, the type of mathematics used in these assessments are representative of the type of linear algebra or graph-theoretic mathematics often used in architecture analysis across multiple metrics.…”

Towards a solver-aware systems architecting framework: leveraging experts, specialists and the crowd to design innovative complex systems

“…Fundamental to the design of complex systems is the core organisational function of coordinating interdependent tasks (cf., Galbraith 1974; Thompson 2003). Interdependent tasks arise when complex systems are partitioned into lower complexity subproblems (Simon 1962, 1996).…”

“…However, since most complex problems are only partially decomposable (Simon 1962), there remains a critical task of managing those interdependencies (Campagnolo & Camuffo 2010). This led organisational scholars to conceptualise design as an organisational problem-solving process where the goal is to place organisational links such that scarce cognitive resources are conserved (Baldwin & Clark 2000; Colfer & Baldwin 2016) and to build theory around where such ties should be placed (Parnas 1972; Hoffman & Weiss 2001; Thompson 2003).…”

“…Overall, the literature recognises decomposition (through modularity or otherwise) as a key strategy in complexity reduction and management, and to enable sustained value across long uncertain lifetimes. However, there is also recognition that too much decomposition can be detrimental to system performance (Ethiraj & Levinthal 2004;Topcu et al 2021). The need to identify the right balance is particularly poignant when considering contributions from nonexpert solvers, as in open innovation.…”