WICSA 2009 WS8 -- Adaptive Architectures

From WICSA Conference Wiki

This page was written around the time of WICSA 2009.

Session Chairs: TBD

Wednesday, 16th September, 13:30 - 15:30

Contents 1 Participants 2 Papers 3 Contributions 4 Working Session Details 4.1 Goal 4.2 Strategy 4.3 Results

Participants

Tell us something about your background. Add a few sentences about the working session topic such as your position, questions you would like to see discussed, etc.

• Judith Stafford, Tufts University. What is architecture anyway? If the system is allowed to change at will, maybe new types of components come into being and join the system, is it possible to talk about the system's architecture?
• Tomi Männistö, Helsinki University of technology. What about the kind of changes to adapt to?
• Carlos E. Cuesta, Rey Juan Carlos University, Spain.
• Christelle Urtado, LGI2P / Ecole des Mines d'Ales
• Eila Ovaska, VTT Technical Research Centre of Finland
• Claudia Raibulet, University of Milano Bicocca, Italy
• Vivek Nallur, University of Birmingham, United Kingdom
• Bradley Schmerl, Carnegie Mellon University, USA
• Romina Spalazzese, University of L'Aquila, Italy
• Sam Malek, George Mason University, DC, USA
• Antonio Bucchiarone, FBK-IRST, Trento, Italy
• Patrizio Pelliccione, University of L'Aquila,
• Rami Bahsoon, The University of Birmingham, UK
• Ismael Bouassida Rodriguez,LAAS CNRS,France
• Rogerio De Lemos, University of Coimbra, Portugal
• Arjan de Roo, University of Twente, the Netherlands

Papers

• "Evaluation of Dynamic Adaptivity Through Metrics: an Achievable Target?", by Claudia Raibulet and Laura Masciadri

  Dynamic adaptivity is one of the most challenging non-functional issues to address during the engineering of information systems. Furthermore, it is even harder to evaluate. The definition of mechanisms through which adaptivity may be evaluated would be very useful both for the maintainability, reusability, and evolution of adaptive systems, and for the relations which may be established among adaptive systems (e.g., interaction, integration). In this paper we propose examples of metrics for dynamic adaptivity which may be considered a starting point for future research on (1) the feasibility related to the definition of measurable evaluation mechanisms for adaptive systems, and (2) the usability of these metrics as design hints in the development process of new adaptive systems and as formal approaches for the evaluation of existing adaptive systems.

• "Towards a Formalization of Mediating Connectors for on the Fly Interoperability", by Romina Spalazzese, Paola Inverardi and Valérie Issarny

  Mediators stand as a core architectural paradigm for today's and future systems that increasingly need to be connected. The mediator concept has been used to cope with many heterogeneity dimensions. Still, a key challenge for today's systems architectures is to embed the necessary support for automated mediation, i.e., the connector concept needs to evolve towards the one of mediating connector. In this paper, we introduce a framework to formalize mediating connectors. The proposed characterization paves the way for automated reasoning about protocol matching and mapping, and thus for the dynamic synthesis of mediating connectors to enable eternal networked systems, which we investigate as part of the CONNECT European project.

• "An Architectural Style for Optimizing System Qualities in Adaptive Embedded Systems Using Multi-Objective Optimization", by Arjan de Roo, Hasan Sözer and Mehmet Akşit

  Customers of today's complex embedded systems demand the optimization of multiple system qualities under varying operational conditions. To be able to influence the system qualities, the system must have parameters that can be adapted. Constraints may be defined on the value of these parameters. Optimizing multiple system qualities under the given set of parameters and constraints is called Multi-Objective Optimization (MOO). This is a well-known mathematical problem, for which numerous solutions have been proposed. The application of an MOO solution in an embedded system involves specific design decisions. It is preferable that these design decisions are documented in the architectural description. Therefore, this paper presents an architectural style, which specializes the Component-and-Connector viewtype, to enable the analysis and design of an architecture from an MOO point of view. A case study from industry is used to demonstrate the usage of this style.

• "A Model-Based Multi-Level Architectural Reconfiguration Applied to Adaptability Management in Context-Aware Cooperative Communication Support Systems", by Ismael Bouassida Rodriguez, Khalil Drira, Christophe Chassot and Mohamed Jmaiel

  Handling context-aware dynamically adaptable architectures contributes to the design of self-configuring software systems. Dealing with such a problem for communicating systems is even more challenging since adaptation should address simultaneously the different communication levels. In this paper, we address this problem by providing a modelbased, rule-oriented approach that supports the adaptation process based on a run-time transformation of the system architecture. Such an architecture may represent the different possible service compositions and the associated architectural configurations. Adaptation policies are defined by means of SWRL rules. Such rules allow associating the suitable adaptation transformations to the context change.

Contributions

If you have relevant materials or references, please add here:

• What Wikipedia has to say about Adaptive Systems
• ADAPTIVE 2009
• Usenix LISA Conference -- LISA papers deal with issues relevant to system administration and often include work on dynamic adaptation of networked computers.

Working Session Details

Welcome to the working session on Adaptive Architectures. The guiding principle for this working session is quite straightforward, architecture-based tools and techniques can be leveraged to assist software systems in making adaptations during runtime. As a means of introduction, the working session will be framed around the ideas presented in the papers, each of which will be presented very briefly at the beginning of the session. For those who attend but do not have papers, you can also take a few minutes to present your thoughts to the group. Once peoples thoughts and concerns have been stated the group will decide on one to three related questions for which answers will be sought during the working session.

To help us get started,include here just one question you hope to find an answer to during this session.

Goal

The goal of this session is to address issues encountered in the research presented by speakers and to identify opportunities for future research. We recognize that there are a variety of axis of variability. We will look for identifying strength and limitations in applying the presented research along these different types of adaptivity.

Strategy

Each participant was encouraged to make a statement about their primary concern with adaptive architectures. First, the accepted papers were presented by their authors in a brief 10 minute talk. Once all talks were completed each of the other attendees introduces him/herself including a brief statement of their concern with adaptive architectures. After these introductions the speakers formed a panel and address questions by other participants. The question/answer period was followed by a general discussion session with the goal of identifying strengths, compatabilities, and limitations of applicability of the approaches presented by speakers in terms of type of adaptability and/or type of system.

Results

The working session was quite lively with nearly two dozen interested persons, some from academia, some from industry, some from industry and academia at the same time. Some attendees have been working on self-adaptive systems for years and for others this was a relatively new topic. The discussion ranged over a variety of topics but resulted in recognition that there are different kinds of variability and variability that happens at different times in a systems life. Some is planned with knowledge of all possible components and configurations pre-defined. Others is ad hoc with components of being added and subtracted from the system at run-time. While these differences are major it was observed that each of the four techniques presented by authors would apply across the situations. However, it was noted that the application of static techniques to the more dynamic situations would require significant resources and will not be possible in the near future. One major concern with all of the work achieving self-adaptation through application of utility functions is the fact that we don’t yet know how to get the system user to describe what they really want and until we can resolve that question the use of any of these techniques can guarantee their use will improve he likelihood that the resulting system will be optimal in the eyes of user.

To summarize the author panel’s question/answer period, the main thrusts of questions were as follows:

Claudia, who described a set of metrics for adaptivity, fielded questions about what point in system life would measures for her metrics would be determined. It was recognized that the metrics could be used at a variety of points in the lifecycle but the intent is to use them statically.

Claudia was asked how she plans to validate her metrics? Did she plan to experiment with long-lived software? Her plan is to continue to collect case studies but also hopes to get access to real systems since case studies from literature studies tend to be designed to be adapted in certain ways and do not really represent reality.

Romina described automatic generation of mediating connectors, which allow new components to come dynamically and be automatically included in the system. One participant asked her if she didn’t need some kind of meta-protocol to start the matching. She answered that she do an abstraction of the behavior of the components and that she assume to have the ontology describing the components behavior and an ontology mapping between ontologies. Another participant observed that the ontology is the meta-model. In response to that Ismael asked her how she populates the ontology mapping, to which she responded that till now she is assuming to have it. But it is possible to populate it in a collaborative way. Bradley asked whether the protocol matching can be done at run-time or needs to be done off-line. She answered that for now is done off-line but one of the future work is to do it at run time (investigated as part of the CONNECT European Project http://www.connect-forever.eu/).

For Arjan presented a method to support making tradeoffs between various optional reconfigurations using MOO, multi-objective functions to perform optimal reconfiguration during run-time in response to changes in system environment. Discussion ensued about the degree of environmental change that is allowed. Arjan explained that change in any dimension is assumed to be within some reasonable limit. The set of components will be predefined and the components themselves are adapted at run-time. Other questions were asked about adding new variables and new objectives, about priority setting. It was noted that optimizing qualities is very subjective, in response Arjan noted that the user is allowed to modify preferences, which raised issues around the possibility that the user would specify they want everything. Arjan noted that this is not possible and user will be forced to prioritize.

Ismael described a self-reconfiguring system that is constrained by transformation rules. In response to a questions about dynamism he noted that his transformations can be applied in either case. The graph grammer can also be used at any level. A question was raised about whether the system would respond to user requests. Ismael responded that the system chooses the graph that is needed and deploys the architecture.