SimCA vs ActivFORMS: comparing control- and architecture-based adaptation on the TAS exemplar

Abstract

Today customers require software systems to provide particular levels of qualities, while operating under dynamically changing conditions. These requirements can be met with di↵erent self-adaptation approaches. Recently, we developed two approaches that are di↵erent in nature -control theorybased SimCA and architecture-based ActivFORMS -to endow software systems with self-adaptation, providing guarantees on desired behavior. However, it is unclear which of the two approaches should be used in di↵erent adaptation scenarios and how e↵ective they are in comparison to each other. In this paper, we apply SimCA and ActivFORMS to the Tele Assistance System (TAS) exemplar and compare obtained results, demonstrating the di↵erence in achieved qualities and formal guarantees.

Control-Theoretical Software Adaptation: A Systematic Literature Review

“…For example, when we found a conference and a journal version of the same paper, we kept only the journal version, as it is considered more complete and accurate. The final list of primary studies for our literature review consists of the following 42 references: [21], [24], [27], [28], [29], [34], [70], [74], [75], [76], [77], [78], [79], [80], [81], [82], [83], [84], [85], [86], [87], [88], [89], [90], [91], [92], [93], [94], [95], [96], [97], [98], [99], [100], [101], [102], [103], [104], [105], [106], [107], [108]. 5.…”

“…Software, in this case, was used as an application domain, typical software engineering aspects were not well elaborated. The remaining 21 studies [21], [24], [29], [34], [70], [76], [79], [80], [81], [82], [83], [88], [92], [94], [96], [98], [99], [104], [105], [106], [107] took an integrated perspective. These studies employed both software engineering and control theoretic aspects to realize adaptation of software.…”

“…Linear model Learned at Runtime (LLR) [21], [29], [34], [70], [74], [75], [76], [77], [81], [82] Model of the form:…”

“…, where u(k + 1) is the system output, η(k) is the actuation signal, α is a coefficient, d(k) is a disturbance acting on the output. u(k + 1) and η(k) are cases-specific; coefficient α is calculated during system identification based on a series of experiments [21], [70], [82] or using a controller [34], [81]. In some cases [21], [82] d(k) is removed from the model, while α is updated during system operation to cope with system dynamics.…”

Towards flexible and secure end-to-end communication in industry 4.0

“…The FEDerAteS (A Foundation for Engineering Decentralized Self-Adaptive Software Systems) project has the overall goal to study and develop a scientific foundation for engineering decentralized self-adaptive systems. The main project results include foundations of self-adaptation, based on which they have provided evidence that external Feedback Control Loops (FCL) improve the design of self-adaptive systems [11], an architecture framework for collective intelligent systems that include humans in the loop [12], and validation of the results in different application domains, such as robotic system [13].…”