Spécification et génération de tests du comportement dynamique des systèmes à objets répartis

Mazziotta, Sandro
Thesis

Modeling in the context of distributed object frameworks is a difficult problem because it requires the integration of three research areas : (1) object oriented modeling, (2) formalization, and (3) distributed processing. The open distributed processing (ODP) reference model defines the theoretical framework to realize this integration. This thesis proposes an original approach to this problem through a pragmatic instantiation of this integration in the perspective of ODP.
By deliberately limiting the scope of the work to functional models, and to information and computational viewpoint issues of the ODP reference model, it is shown how a very abstract and expressive behavior specification framework can be devised. The specification framework is based on a declarative specification of actions. Actions in the system are declared with respect to all the functionalities to be modeled. The behavior specification template is a variant of the well known event-condition-action (ECA) rule model, extended by the specification of safety properties or assertions, that are used during validation. With a declarative approach control issues in the system that typically overspecify a functional model can be abstracted away to a maximal extent. In addition, the specification framework makes possible the integration of any high level aggregation structures typically proposed by most of the object oriented analysis and design methods. In particular, roles and relationships are used to provide very expressive modeling. Note well that the specification framework is generic in the sense that it does not depend on any particular distributed computing notation. In particular both OSI systems management (GDMO, ASN.1) and CORBA (IDL) notations can be integrated. Each element specified in such notations is integrated according to the role played with respect to either the information or the computational ODP viewpoint. Validation is based on execution, therefore an important part of the thesis consists of defining how the declarative specification of actions model can be made executable. To this end a precise operational semantics is defined and implemented by an algorithm called the behavior propagation engine (BPE) algorithm. Since the proposed declarative specification framework follows the very general ECA-rule model, the execution semantics can interestingly be based on principles already stated in the context of active database management systems (ADBMS). The result is a transition system, defined by a set of transition functions that exercise their processing on a user-level control structure called the behavior execution tree (BET). The BET gives a complete representation of the control state at each execution step. The BET can be visualized and be used for user interaction and debugging purposes. In addition, execution backtracking is a powerful problem analysis facility that has been incorporated in the interactive execution environment. This facility is supported by undo functions and the use of continuations. Finally, the validation environment incorporates reduced state space exploration algorithms based on a combination of partial order and state caching techniques. Jointly, interactive execution and state space exploration are very complementary validation tools. State space exploration is typically used to detect problems in a model, and interactive execution performs very well for problem analysis. The concrete result of this thesis is a powerful tool-set that is in addition expected to be usable by distributed object computing systems engineers in their daily work in specification. A tool allows specifiers to get immediate and concrete feedback about their work. In addition, engineers are forced to get a detailed knowledge of the models considered. Though a significant amount of work may be needed to get a model complete, the result is a better understanding of models. Finally, the corresponding specification can be implemented faster. The tool-set was developed in the context of the TIMS project. TIMS stands for TMN-based Information Model Simulator. It is a joint project between "Institut Eur´ ecom" and "Swisscom". In the thesis references are made about the application of the tool-set in the context of the TIMS project.


Type:
Thèse
Date:
1997-10-24
Department:
Sécurité numérique
Eurecom Ref:
903
Copyright:
© Université de Nice. Personal use of this material is permitted. The definitive version of this paper was published in Thesis and is available at :
See also:

PERMALINK : https://www.eurecom.fr/publication/903