Munich/MIT Survey: Development of Embedded Linux
by Joachim Henkel (Munich, Leipzig) and Mark Tins
FULL TEXT
Introduction
Much of the early research on Free/Open Source software (FOSS) has focused on the activity and personal motivations of individual contributors. This has led to a popular view that FOSS is a fringe activity pursued by hobbyists, antithetical to the work of profit-driven firms. Increasingly, however, evidence is accumulating that, although hobbyists and personal motivations are important, a very large part of the development effort is conducted by programmers at work, with the knowledge of their supervisors (see Lakhani and Wolf).
In other words, firms are choosing to participate in FOSS development. Although some of these firms may contribute for strategic reasons (for example, IBM’s Linux efforts undercut rival Microsoft), the software plays no strategic role for other firms: few of the firms contributing to Linux are direct Microsoft competitors. One important group of firms develops “embedded Linux”—variants of Linux adapted, in various ways, to the use in hardware devices. For many of these firms, FOSS is an integral part of the products they offer, yet they freely share much of their code with other developers, including competitors. Despite this practice—or perhaps because of it—embedded Linux has rapidly become a very successful product, adopted by many device manufacturers as well as specialized software firms, and has competed successfully against embedded versions of Microsoft Windows and Wind River’s VxWorks.
Survey
To understand the nature and motivation of these firms, Joachim Henkel and Mark Tins conducted a survey of embedded Linux developers. The survey was conducted through Web sites and email and had 268 valid (self-selected) respondents. The authors note that this self-selection may mean that hobbyists are over-represented in the sample.
Nevertheless, hobbyists comprised only 15 percent of the sample; employees of device manufacturers (42 percent), software firms specializing in embedded operating systems (22 percent) and component manufacturers (9 percent) made up the majority. The remainder consisted of people at universities and non-profit organizations. Many of the developers’ employers were small companies, but about one third of the sample worked at firms with more than 200 employees.
On average, respondents working for commercial firms spent 28 hours per week at work on embedded Linux, and roughly 7 hours in their spare time. Over 30 percent of the sample worked more than 38 hours per week at work. Clearly, although hobbyists and leisure-time activity contribute a significant fraction of the effort, most effort is done at work by employees. Moreover, these developers are hardly amateurs—mean experience with software for the sample is 14 years.
Sharing Code
A key set of questions asked developers to what extent firms shared code that was useful to others (that is, “not too specific” to the firm), and why. Software companies shared 57 percent of such code, and hardware companies shared 45 percent. In many cases, the shared code was specific to the firm’s hardware and software, but much was also generic (53 percent for hardware companies, 70 percent for software). In addition, firms shared code even in cases where the code helped to differentiate them from others or was important to their competitive position.
The top five reasons software firms gave for sharing code were: 1) the GPL license requires it; 2) other developers fix bugs and share these fixes; 3) the firm wants to appear as a good member of the open source community; 4.) it reduces their maintenance burden when their modifications become part of the standard distribution; and 5) others improve the code further and share their improvements in turn. The survey also asked a series of questions about code that is not shared and found that firms keep some portion of code secret for a variety of reasons and using a variety of methods. The author further explores issues about sharing in a companion paper, Patterns of Free Revealing.
The survey also looked in-depth at the frequency of developers’ participation in different activities supporting FOSS projects, the nature of communications with other developers outside their own firm, factors that hinder FOSS development, and developers’ identification with the broader Free/Open Source software community. The overall picture that emerges is of an industry where developers participate in a variety of ways and at different levels, but where practical economic benefits play a dominant role in generating participation by firms.
