| Abstract |
In emerging domains such as ubiquitous computing, service robotics, unmanned space and water exploration, and medical and life-support devices, software is becoming increasingly complex with extensive variation in both requirements and resource constraints. Developers face growing pressure to deliver high-quality software with additional functionality, on tight deadlines, and more economically. In addition, modern computing and network environments demand a higher degree of adaptability from their software systems. Computing environments, user requirements, and interface mechanisms between software and hardware devices such as sensors can change dynamically during runtime. Because it's impossible to foresee all the functionality or variability an SPL requires, there is a need for dynamic SPLs that produce software capable of adapting to fluctuations in user needs and evolving resource constraints. DSPLs bind variation points at runtime, initially when software is launched, to adapt to the current environment, as well as during operation to adapt to changes in the environment. Although traditional SPL engineering recognizes that variation points are bound at different stages of development, and possibly also at runtime, it typically binds variation points before delivery of the software. In contrast, DSPL engineers typically aren't concerned with pre-runtime variation points. However, they recognize that in practice mixed approaches might be viable, where some variation points related to the environment's static properties are bound before runtime and others related to the dynamic properties are bound at runtime. In DSPLs, monitoring the current situation and controlling the adaptation are thus central tasks. The user, the application, or generic middleware can perform these tasks manually or automatically. Although dynamic software product lines build on the central ideas of SPLs, there are also differences. For example, the focus on understanding the market and letting the SPL drive variability analysis is less relevant to DSPLs, whose primary goal is to adapt to variations in individual needs and situations rather than market forces. In summary, a DSPL has many, if not all, of the following properties: dynamic variability configuration and binding at runtime, changes binding several times during its lifetime, variation points change during runtime: variation point addition (by extending one variation point), deals with unexpected changes (in some limited way), deals with changes by users, such as functional or quality requirements, context awareness (optional) and situation awareness, autonomic or self-adaptive properties (optional), automatic decision making (optional), and individual environment/context situation instead of a "market." Given these characteristics, DSPLs benefits from research in several related areas. For example, situation monitoring and adaptive decision making are also characteristics of autonomic computing, and the DSPL approach can be seen as one among several to building self-adapting/managing/healing systems. In addition, dynamically reconfigurable architectures provide mechanisms to rebind variation points at runtime, while multiagent systems, which focus on the use of agents and communities of agents, are particularly useful for evolving systems such as DSPLs [1]. |
