Policy-based management is an approach in which organization guidelines can be expressed as policies that are enforced by a management system. These rules specify the corrective actions that should be executed in different situations and are designed using the Event-Condition-Action (ECA) rule framework. The events and conditions express the situation in which the corrective action should be executed.A management policy evolves over time by addition and removal of rules, policy composition, rule modifications and due to various system dynamisms. These changes may result in conflicts and cycles among rules. Multiple rules can become simultaneously eligible for enforcement in a situation and the order of enforcement may determine the final system state. Rule enforcement may fail necessitating an exception model for policies. Rules may contain long-running actions, which may cause conflicts with rules that are enforced at a later situation. The dynamism of pervasive systems with frequently changing configurations further complicates policy design. In order to address these issues, complex static and dynamic reasoning techniques have to be supported by management systems. The ECA rule framework is poorly suited for designing management policies for pervasive systems since it does not contain any information about the rule action. A rule action is initiated on the specified situation and there is no information about the effect of the action on the system or whether the action completed execution successfully. This information is vital for policy analysis and for providing various guarantees. The above problems lead to non-determinism, which makes the management process unpredictable.In this thesis, we propose a rule framework called Event-Condition-Precondition-Action-Postcondition (ECPAP) that contains axiomatic specifications of rule actions, for designing management policies. These specifications formally state the effect of an action using Hoare logic as pre- and post-conditions. This framework facilitates advanced conflict and cycle analysis, determines enforcement order when multiple rules are simultaneously triggered, supports policy exception handling and provides reasoning support for rules with long-running actions. We show how the ECPAP framework enables deterministic policy-based management.We propose algorithms for static and dynamic analyses, enforcement verification and monitoring and reasoning with long-running actions. We demonstrate the need for these algorithms on various distributed and pervasive systems and evaluate their performance. Our experiments show that the ECPAP framework leads to effective policy-based management and is a feasible approach.
展开▼
