An Empirical Comparison of Reuse in Embedded and Nonembedded Systems

High-quality software, delivered on time and budget, constitutes a critical part of most products and services in modern society. Our government has invested billions of dollars to develop software assets, often to redevelop the same capability many times. Recognizing the waste involved in redeveloping these assets, in 1992 the Department of Defense issued the Software Reuse Initiative. The vision of the Software Reuse Initiative was "To drive the DoD software community from its current "re-invent the software" cycle to a process-driven, domain-specific, architecture-centric, library-based way of constructing software.'' Twenty years after issuing this initiative, there is evidence of this vision beginning to be realized in nonembedded systems. However, virtually every large embedded system undertaken has incurred large cost and schedule overruns. Investigations into the root cause of these overruns implicates reuse. Why are we seeing improvements in the outcomes of these large scale nonembedded systems and worse outcomes in embedded systems? This question is the foundation for this research. The experiences of the Aerospace industry have led to a number of questions about reuse and how the industry is employing reuse in embedded systems. For example, does reuse in embedded systems yield the same outcomes as in nonembedded systems? Are the outcomes positive? If the outcomes are different, it may indicate that embedded systems should not use data from nonembedded systems for estimation. Are embedded systems using the same development approaches as nonembedded systems? Does the development approach make a difference? If embedded systems develop software differently from nonembedded systems, it may mean that the same processes do not apply to both types of systems. What about the reuse of different artifacts? Perhaps there are certain artifacts that, when reused, contribute more or are more difficult to use in embedded systems. Finally, what are the success factors and obstacles to reuse? Are they the same in embedded systems as in nonembedded systems? The research in this dissertation is comprised of a series of empirical studies using professionals in the aerospace and defense industry as its subjects. The main focus has been to investigate the reuse practices of embedded systems professionals and nonembedded systems professionals and compare the methods and artifacts used against the outcomes. The research has followed a combined qualitative and quantitative design approach. The qualitative data were collected by surveying software and systems engineers, interviewing senior developers, and reading numerous documents and other studies. Quantitative data were derived from converting survey and interview respondents' answers into coding that could be counted and measured. From the search of existing empirical literature, we learned that reuse in embedded systems are in fact significantly different from nonembedded systems, particularly in effort in model based development approach and quality where the development approach was not specified. The questionnaire showed differences in the development approach used in embedded projects from nonembedded projects, in particular, embedded systems were significantly more likely to use a heritage/legacy development approach. There was also a difference in the artifacts used, with embedded systems more likely to reuse hardware, test products, and test clusters. Nearly all the projects reported using code, but the questionnaire showed that the reuse of code brought mixed results. One of the differences expressed by the respondents to the questionnaire was the difficulty in reuse of code for embedded systems when the platform changed. The semistructured interviews were performed to tell us why the phenomena in the review of literature and the questionnaire were observed. We asked respected industry professionals, such as senior fellows, fellows and distinguished members of technical staff, about their experiences with reuse. We learned that many embedded systems used heritage/legacy development approaches because their systems had been around for many years, before models and modeling tools became available. We learned that reuse of code is beneficial primarily when the code does not require modification, but, especially in embedded systems, once it has to be changed, reuse of code yields few benefits. Finally, while platform independence is a goal for many in nonembedded systems, it is certainly not a goal for the embedded systems professionals and in many cases it is a detriment. However, both embedded and nonembedded systems professionals endorsed the idea of platform standardization. Finally, we conclude that while reuse in embedded systems and nonembedded systems is different today, they are converging. As heritage embedded systems are phased out, models become more robust and platforms are standardized, reuse in embedded systems will become more like nonembedded systems.

Comparison of Reading Development Across Socioeconomic Status in the United States

It is well known that higher parental socioeconomic status (SES) predicts better child reading outcomes, but little work has been done to unpack this finding. The main overall question addressed by this project was whether cognitive models of the two main reading outcomes, single word reading (SWR) and reading comprehension (RC), performed similarly across levels of parental SES. The current study predicted a differential relation between parental SES and both predictors and outcomes because of the known large relation between parental SES and child oral language development. Three questions examined the mediating effects of cognitive predictors on the relation between parental SES and reading outcomes, the moderating effects of SES on the developmental trajectories of reading outcomes, and the strength of the relationship between SES and the two reading outcomes. Participants were part of two large and comprehensive datasets: the cross-sectional Colorado Learning Disability Research Center (CLDRC; n=1554) sample, and the International Longitudinal Twin Study (ILTS; n=463 twin pairs) sample. In terms of cognitive predictors, the relation between SES and SWR was disproportionately mediated by two language skills, vocabulary (VOC) and phonological awareness (PA). For the RC models, both SWR and oral listening comprehension (OLC) did not disproportionally mediate the relation between RC and SES; however, full mediation was not exhibited. With regard to the trajectory of reading outcomes, SES moderated the starting values of SWR and RC, and the slopes of SWR development. When performance on the control measures of early reading skills (e.g., print knowledge, vocabulary, and decoding skills) was included the models, the moderating effects of SES were completely accounted for by these measures. In terms of outcomes, SES had a stronger relation to RC than to SWR, especially at later ages. These findings have implications for interventions aimed at improving reading outcomes in children from lower SES families.