Konfliktbehandlung im policy-basierten Management mittels A-priori-Modellierung

Policy based management plays an important role in the research community and in the industry. Policies can be in conflict to each other due to specification of policies by several persons or when targeting divergent goals. Existing approaches for policy detection and resolution are limited: they are only applicable for distinct policy languages. They are only able to detect certain conflict types or offer no methodology to integrate handling of new conflicts types. The assessment of conflicts shows that by analyzing management models, new kinds of conflicts can be identified, which are not addressed by existing solutions in a generic sense. Therefore management models are used as a priori models. An a priori model describes a target state of a system and can be seen as set of conditions a policy system has to observe. Under this premise a new conflict type — a conflict between related managed objects — can be identified. The core of the solution presented is a methodology to derive conflict definitions out of (a priori) model properties. Starting with the definition of invariants of model properties, policy actions are linked with these invariants and in the end preconditions are specified which prevent policy conflicts from occurring. The general applicability of the methodology is demonstrated for functional dependencies , containment hierarchies and finite automata. A new algorithm was developed for conflict handling, which consisting of the phases conflict localization, conflict detection and conflict solution. In the first phase, the algorithm quickly reduces the set of policies to be considered. After having performed conflict detection by means of invariant evaluation in the second phase, in the last phase, for each conflict type a distinct strategy is developed to ensure optimal conflict solution. The algorithm is applicable both for preventive and reactive conflict handling. To achieve a generic solution, important design goals for the methodology and the algorithm were independence of specific policy languages, support for conflict types not known at design time and independence of a specific management information model. The applicability of the approach in practice is demonstrated in an exemplary manner by the mapping of conflict definitions to the Common Information Model.