Architecture-driven Self-adaptation of Mixed Systems

Over the past decades we have observed a trend towards large-scale systems. These systems are characterized by decentralized control, continuous change of elements, omnipresence of failures, and simultaneous presence of conflicting interests and collaborative behavior. The boundary between humans and software becomes blurry as humans become providers and consumers of content and computation. In such mixed systems, the overall interaction topology of humans and software entities has a profound effect on the system`s run-time management. First, relations between humans and software become complex due to the system`s scale and entity dynamics. Thus, focusing adaptation on individual elements only cannot yield successful results. Second, entities in large-scale systems can only obtain a local view of the overall system. Subsequently, necessary global level changes remain unexplored by individual entities. These two factors outline the need for self-adaptation. To this end, however we need to address following challenges: " How to describe optimal and critical configurations of large-scale systems? " How to measure a mixed system`s underlying capability to adapt? " How can bio-inspired, self-organizing adaption mechanisms be applied in mixed systems? The ultimate goal of the research program is providing models, metrics, and techniques to: " Understand the configuration of a mixed system. " Understand what adaptation can happen. " Understand how adaptation can happen. " Understand why adaptation has happened. The proposed research agenda takes architectural styles as the fundamental engineering principle and combines them with models and metrics of large-scale mixed systems. The research methodology consists of the following five tasks: 1. The central hypothesis assumes that different architectural styles need to be applied to the software entity architecture and the architecture of the human interaction support subsystem. This requires an initial assessment of existing architectural styles with respect to their adaptation impact in mixed systems. 2. Graph-centric interaction metrics from the domain of Social Network Analysis provide the basis for describing a mixed system`s configuration. Additional similarity metrics enable the comparison of desired and actual system status. 3. Aggregation of QoS-like metrics of individual entities, graph-centric interaction metrics, and available adaptation actions will enable qualitative metrics that describe a system`s adaptivity. 4. The grounding of self-organizing algorithms requires the modeling of adaptation policies that apply local information only. The appropriate policy (addressing human and software entities separately) depends on the underlying architectural style. 5. Adaptation tracing requires the identification of systems states with similar behavioral characteristics. Clustering enables the aggregation of similar, frequently reoccurring system configurations and subsequently allows extraction and interpretation of a system`s change transition graph.