DoD Software Summit,2001 Strategic Software Design Kevin Sullivan University of Virginia Department of Computer Science
DoD Software Summit,2001 Overview Value added is overriding goal of design How do technical choices create value? Missing even basic science in this area Objective: enable design for value added
DoD Software Summit,2001 Example Still don’t understand modularity well enough What is it? –design structure matrix [Ulrich, Eppinger] –design rules [Baldwin, Clark] How does it add value? –real options to search design space –decentralized control over search
DoD Software Summit,2001 DSM: Non-Modular Design Design parameters (and design tasks) Parameter interdependencies (implied communications)
DoD Software Summit,2001 DSM: Proto-Modular Design Proto-modules (and corresponding aggregate tasks)
DoD Software Summit,2001 DSM: Modular Design design rule constrains search decouples taskscreating independent modules, and valuable substitution options
DoD Software Summit,2001 DSM in Environment (EDSM) Sub-Optimal Modular Design Design rules sensitive to change in environment
DoD Software Summit,2001 EDSM for High Value Modularity Design rules invariant under change in environment
DoD Software Summit,2001 How Does It Add Value? Overall value = present value of function in environment + present value of options to improve/adapt Module-level option value (substitution case) expected benefit: best of k candidate replacements – cost of creating them – cost of ripple effects IBM360: 25X in options value! [B&C 2000]
DoD Software Summit,2001 Concluding Comments Aim for generalized concept of modularity well beyond inadequate OO model E.g., to aspect-oriented modularity power, real time response, hardware … Options valuation model certainly not proven for production use sufficient to capture seminal results [Parnas72] Insights into drivers of design & industry dynamics Exploring tool support –Make DSMs scale with hierarchical design parameters
DoD Software Summit,2001
Parnas 72: Traditional Modularity FUNCTION SIGNATURES DATA STRUCTURES ALGORITHMS
DoD Software Summit,2001 Parnas 72: IH Modularity ADT INTERFACES DATA STRUCTURES & ALGORITHMS
DoD Software Summit,2001 EDSM: Traditional Modularity ENVIRONMENT PARAMETERS IMPACT OF EPs ON DESIGN RULES AND ON INDEPENDENT MODULES
DoD Software Summit,2001 EDSM: IH Modularity INVARIANCE OF DESIGN RULES UNDER ENVIRONMENTAL CHANGE
DoD Software Summit,2001 Comparative Net Options Value NOV = E (Benefit of k Experiments) – Cost of k Experiments – Ripple Costs
DoD Software Summit,2001 Parnas 72: Non-Modular Design IN SHIFT SORT OUT CTL