Karim Guennoun.

Dynamic architectures in the context of component-based and service-oriented Applications. PAUL SABATIER University, Toulouse, FRANCE. LAAS Rapport N06903

Abstract: Adaptability for software applications can be separated into two categories. First relates to the behavioral adaptation also called algorithmic adaptation. This adaptation addresses the redefinition of the behavior of the application and its components and implies, for example, adding a new method into the interface of a component or updating the orchestration protocol that coordinates a set of services. Second category, in which we can classify our work, relates to the structural adaptation and implies a reconfiguration at the architectural level. This kind of reconfiguration deals with the organization of the architecture and involves, for instance, the substitution of a failing component by another with same functionalities or the redirection for a customer of a service which does not respect the QoS contract towards a service likely to offer better guarantees. In this thesis, we specify a meta-model relating to the description and the automatic management of dynamic architectures. Architecture instances are described by extended graphs where components (or services) are represented by vertices, and interdependencies (e.g. connections, relations of control ... etc) are described by edges. The architectural styles are specified by extended graph grammars. The meta-model considers descriptions by admitting various abstraction levels and others mechanisms to either abstract or refine descriptions according to specific points of view. It, also, makes it possible to describe architecture management protocols and to characterize the architectural properties to be preserved for each considered architectural level. We developed an algorithm for graph homomorphisms building and an algorithm for graph transformation in the context of extended graph grammars defined for our meta-model. Complexity analysis of these algorithms as well as the experimental results obtained made it possible to show their effectiveness. A second version of the two algorithms was defined benefiting from the specificity of our graph transformation context. Complexity analysis of these new versions gives results even more powerful when considering their scalability. Our approach was applied to a case study in the context of the activities of emergency operations. The related system implies structured groups of robots or soldiers that have unequal resources for communication capacities, CPU and energy. The needs for adaptability rise from the changes in the execution context, from fault occurrence, and to achieve QoS provisioning.


Dynamic architectural reconfiguration using a graph transformation. An experimental study of the scalability

Rapport LAAS N08059, Mars 2008, 14p.

Abstract: Architectural adaptation requires complex mechanisms for its management and is important for handling self-configuring properties of autonomic distributed systems. Describing dynamic architectures includes both characterizing conformance of an architecture to a given style description as well as defining rules for architecture evolving through reconfiguration management. Within this research context, several works have been conducted using formal specification to handle this complexity. Graph and graph rewriting-based approaches showed, through many studies, their appropriateness to tackle architectural adaptation problems. However, scalability of these approaches remains an open issue and has been rarely explored. In this paper, we investigate this issue.We, first, introduce a graph-based general approach for handling both aspects of dynamic architecture description, and we illustrate it within a scenario of communication and collaboration support in crisis management systems. We elaborate formal models for style and architecture description and management while considering multiple architectural levels. Using the French Grid GRID5000, we expose an experimental study to validate the elaborated models and give results for the related execution times of implemented architecture management protocols.