The Improvement Cycle: Analyzing Our Experience

NASA’s Software Engineering Laboratory (SEL), one of the earliest pioneers in the areas of software process improvement and measurement, has had a significant impact on the software business at NASA Goddard. At the heart of the SEL’s improvement program is a belief that software products can be improved by optimizing the software engineering process used to develop them and a long-term improvement strategy that facilitates small incremental improvements that accumulate into significant gains. As a result of its efforts, the SEL has incrementally reduced development costs by 60%, decreased error rates by 85%, and reduced cycle time by 25%. In this paper, we analyze the SEL’s experiences on three major improvement initiatives to better understand the cyclic nature of the improvement process and to understand why some improvements take much longer than others. Background Since 1976, the Software Engineering Laboratory (SEL) has been dedicated to understanding and improving the way in which one NASA organization, the Flight Dynamics Division (FDD) at Goddard Space Flight Center, develops, maintains, and manages complex flight dynamics systems. It has done this by developing and refining a continual process improvement approach that allows an organization such as the FDD to fine tune its process for its particular domain. Experimental software engineering and measurement play a significant role in this approach. The Software Engineering Laboratory (SEL) is a partnership of NASA Goddard’s Flight Dynamics Division (FDD), its major software contractor, Computer Sciences Corporation (CSC), and the University of Maryland’s (UM) Department of Computer Science. The FDD primarily builds software systems that provide ground-based flight dynamics support for scientific satellites. They fall into two sets: ground systems and simulators. Ground systems are midsize systems that average around 250 thousand source lines of code (KSLOC). Ground system development projects typically last approximately 2 years. Most of the systems have been built in FORTRAN on mainframes, but recent projects contain subsystems written in C and C++ on workstations. The simulators are smaller systems averaging around 60 KSLOC that provide the test data for the ground systems. Simulator development lasts between 1 and 1.5 years. Most of the simulators have been built in Ada on a VAX computer. The project characteristics of these systems are shown in Table 1. The SEL is responsible