Manual administration of complex distributed applications is almost impossible to achieve. On one side, work in autonomic computing focuses on systems that are able to maintain themselves, driven by high-level policies. Such a selfadministration relies on the concept of a control loop. On the other side, modeling is currently used to ease design of complex distributed systems. Nevertheless, at runtime, models remain useless, because they are decoupled from the running system which is subject to dynamic changes. The autonomic computing control loop involves an abstract representation of the system used to analyze the situation and to adapt the application properly. Our proposition, named Distributed Autonomous Component-based ARchitectures (Dacar), introduces models in the control loop. Using adequate models into the control loop, it is possible to design both the distributed systems and their evolution policies, and to execute them. The metamodel suggested in our work mixes both OMG Deployment and Configuration specification and the Event-Condition-Action (ECA) metamodels. This paper treats the different concerns that are present in the control loop and focuses on the concepts of the metamodel that are needed to express entities of the control loop. It also gives an overview of the current Dacar prototype and illustrated it on an ubiquitous application example.
展开▼
