Section: New Results

Requirements-aware Systems for Self-adaptation under Uncertainty

Participants : Romina Torres, Nelly Bencomo, Valérie Issarny, Peter Sawyer.

The development of software-intensive systems is driven by their requirements. Traditional requirements engineering (RE) methods focus on resolving ambiguities in requirements and advocate specifying requirements in sufficient detail so that the implementation can be checked against them for conformance. In an ideal world, this way of thinking can be very effective. Requirements can be specified clearly, updated as necessary, and evolutions of the software design can be made with the requirements in mind.

Increasingly, however, it is not sufficient to fix requirements statically because they will change at runtime as the operating environment changes. Furthermore, as software systems become more pervasive, there is growing uncertainty about the environment and so requirements changes cannot be predicted at design-time. It is considerations such as these that have led to the development of self-adaptive systems (SASs), which have the ability to dynamically and autonomously reconfigure their behavior to respond to changing external conditions.

The key argument of our research is that current software engineering (SE) methods do not support well the kind of dynamic appraisal of requirements needed by a SAS. definition and structure of requirements is lost as requirements are refined into an implementation. Even in cases where requirements monitoring is explicitly included, high-level system requirements must be manually refined into low-level runtime artefacts during the design process so that they can be monitored. There is a lack of approaches supporting for runtime representation, evolution and assessment of requirements. Currently, the approaches mainly assume that it is possible to predefine and envisage the requirements for the total set of target behaviours. Such estimations and beliefs may not be appropriate, if the system is to recover during execution from unforeseen situations, or adapt dynamically to new environmental conditions or to satisfy new requirements that were not foreseen during development. A self-adaptive system is able, at run time, to satisfy new requirements and behaviors. Our research focuses on approaches to support the runtime representation of requirements that will underpin the way a system can reason and assess them during execution.

Our research has been carried out within the research project Marie Curie Fellowship called Requirements-aware Systems (nickname: Requirements@run.time). The research is based on a new paradigm for SE, called requirements-awareness (also known as requirements reflection), in which requirements are reified as runtime entities. Requirements-awareness allows systems to dynamically reason about themselves at the level of the requirements - in much the same way that architectural reflection currently allows runtime reasoning at the level of software. We believe that requirements-awareness (i.e. requirements reflection) will support the development and management of SASs because it will raise the level of discourse at which a software system is able to reflect upon itself.

In the above context, we have been working on the design and implementation of systems with the ability to dynamically observe and reason about their requirements. The results will contribute towards the development of conceptual foundations, engineering techniques, and computing infrastructure for the access and manipulations of runtime abstractions of requirements. Currently, a prototype for the use of runtime goals has been developed. The RELAX language has been proposed to make requirements more tolerant to environmental uncertainty. Design assumptions, called Claims), are applied as markers of uncertainty that document how design assumptions affect goals. Monitoring Claims at runtime has been used to drive self-adaptation. By monitoring Claims during the execution of the systems, their veracity can be tested. If a Claim is falsified, the effect can be propagated to the system's goal model and an alternative (more suitable) means of goal realization will be selected, resulting in dynamic adaptation of the system to a configuration that better satisfies the goals under the prevailing environmental context.