Article Preview
TopIntroduction
Self-Adaptive Systems (SASs) consists of a Managed Subsystem and a Manager Subsystem, named autonomic manager (AM) (Kephart et al., 2003). The first one deals with the domain functionality (Kephart et al., 2003). While, the AM is responsible for keeping such type of systems, which working on the defined QoS level within highly dynamic, heterogeneous, and unpredictable contexts.
To decrease the complexity and to increase the efficiency of the AMs, researchers propose a subdivision into multiple and parallel autonomic entities as a design approach of multi-attributes control of SASs. So, the self-adaptation of SAS should be realized through the interaction of multiple sub-AMs. While, the autonomic behavior is achieved by implementing the MAPE-K reference model.
Different patterns of control loops composition have been proposed by centralizing and/or decentralizing the sub-parts of the Multiple MAPE-K Control Loops (Weyns et al., 2013). Multiple MAPE-K Control Loops can be composed using the appropriate pattern, such as: Parallel Control Loops, Coordinated Parallel Control Loops, and Hierarchic Control Loops (Sylla et al., 2017). However, there is a clear lack of a standard model to compose the different entities of the Control Loops. The potential downsides of the existing patterns are: (i) When using coordination between all or most entities of the loop (similar to the coordinated control pattern), the system may suffer from the high cost of transferring data between different entities. Moreover, the scalability of systems can be difficult in this case, (ii) The total lack of coordination or invalid selections of entities which requires coordination can introduce problems, for example: conflict in the decision (similar to the information sharing pattern), (iii) significant communication overhead when centralizing some entities (like master/slave pattern). Therefore, it is important to know which entities should be centralized and which ones should be decentralized. On the other hand, it is also necessary to know which entities should be coordinated.
The proposed pattern is built upon the enhancement of models proposed in previous studies. In particular, our contribution focuses on the consequences presented by (Weyns et al., 2013) of a selection MAPE pattern. In this paper, authors describe a pattern to compose Multiple MAPE-K loops. It consists of a combination of centralized and decentralized techniques that can provide flexibility for the adaptation cycle. The problem of conflict is usually happened during the Planning and Execution stages. The Plan entity must be centralized to ensure correct and coherent decisions. Also, the Execution entities must be coordinated to ensure the reasonable execution time of multiple adaptation operations. Finally, researchers benefit from the utility of the Knowledge base to coordinate indirectly the Monitoring entities with each other, as well as the Analysis entities.
In this paper, authors propose a refinement pattern of interacting multiple MAPE-K Loops. It has the following characteristics:
- •
Simple composition of Multiple MAPE-K loops without using complex coordination;
- •
General, which is not limited to a specific domain;
- •
Reliable composition, that allows to achieve global adaptation without conflicts.
The authors implement the proposed pattern based on the Fractal component model (Bruneton et al., 2006), the FScript navigation language (David, 2005) for reliable reconfiguration of the Fractal architecture, and the generic Framework for context-aware application WildCat (David et al., 2005). Fractal, FScript, and WildCat are elements of the SAFRAN Framework (David et al., 2006). A case study in the field of EV is presented to validate the proposed pattern.