A method and system are provided for use in administering a complex system, such as a distributed computing ensemble. A model of the system being administered is prepared, preferably during runtime of the invention, by a combination of autodiscovery processes and manual input of information as needed. The model represents not only the resources found in the administered system, but also the service-relationships among those resources. The system administrator also can define elements in the model corresponding to arbitrary groupings of already-existing parts of the model. Software agents, which can be reconfigured, started and terminated as desired during runtime, report changes in state of the managed resources to the model, which updates itself and explores portions of the model adjacent (in terms of the service relationships) to the affected resource(s). Clusters of neighboring state-changes that have a logical relation to each other are grouped together as an alarm, and are preferably represented in a graphical display. Any root-cause event of this type is marked as such, and any portions of the modeled system being (or logically likely to be) affected by the changes are also identified and displayed.
展开▼
