The Disproportionate Scaling of Software Projects

Key Insight

Software project scaling is non-linear; a program 10 times larger (e.g., from 25000 lines to 250000 lines) requires about 30 times the effort, not 10 times. This increased effort includes approximately 25 times more construction and 40 times more architecture and system testing. Similarly, errors also increase disproportionately, with a program 10 times larger typically having roughly 15 times more errors.

Communication paths increase multiplicatively, proportional to the square of the number of people on a team. A 2-person project has 1 communication path, a 5-person project has 10 paths, a 10-person project has 45 paths, and projects with 50 or more programmers can have at least 1200 potential paths. This surge in communication paths leads to more time spent communicating and a greater likelihood of communication mistakes, necessitating formal organizational techniques like documentation to streamline or limit interactions.

Productivity on small projects can be 2 to 3 times higher than on large projects, with a 5 to 10-fold variation from the smallest to the largest projects. For instance, a 1000-line-of-code project might yield 2500-25000 lines per staff-year, while a 1000000-line-of-code project could yield only 700-10000 lines per staff-year. This general trend indicates that larger projects are inherently less efficient per unit of code produced, despite other influencing factors like personnel quality and programming language.