Evolving Adaptable Systems: Potential and Challenges.

摘要: Future software systems will be even longer in service than today (which often is already measured decades). During operation, a system must change many ways, e.g., due to environment changes, functional enhancements, etc. While adaptive cope with changing environments, existing approaches are insufficient address an open-ended world where unpredictable changes may happen that cannot foreseen. Here, we discuss how work (dynamic) product lines, evolution of lines (e.g. [1]) and [2]) can combined improve evolvability systems. Before the systems, first outline concept variability its relationship A key (SPL) notion [4]. model defines differences among valid products as well interdependencies different characteristics SPL. differentiation typically made at development time, more recently approaches, referred dynamic (DSPL) [5], target variation runtime. SPL seen orthogonal dimension engineering, models used capture organize variants. As explained [1], this requires extensions modeling approaches. These required support potential openness evolutionary changes. The term self-adaptive for capable performing their internal or external behavior structure. This done counter environment, threats, other operational challenges like compensating faults [3]. An consists following major parts / subsystems (cf. Figure 1): • Monitor: collects data on current situation (internally externally system) basis decisions system. Adaptivity model: describes adaptations constraints them (i.e., possible modifications effect benefit function). planning adaptations. only explicit implicit control. manager: interprets adaptivity determines It component. executor: realizes adaptation plan, i.e., performs actual Capability implementations realize functionality Adaptation modifies parameters interconnections them. If compare DSPLs, see DSPLs class (runtime) corresponds should noted, however, sometimes not capabilities represented, but directly encoded (removing need component). However, kind information complex typical model. usually logic-based, contain metrics trade-offs. What common both cases scope any point time range operations We call also closed total space (implicitly) predefined. leads shortcoming adaptivity. They able handle runtime, within predefined scope. Long-living extending space, no matter thorough analysis construction done, there always which have been expected. Such i.e. capability currently supported. In next section, basic approach opening provide evolution.