Distributed systems characterized by a high degree of inter-computer resource sharing generally perform better if resources are managed utilizing as much knowledge of the current global state of the system as possible. Decentralized resource management schemes have been preferred over centralized schemes for reasons of reliability, autonomy, speed, and symmetry. Yet, distinct computers in a distributed system often view the global system state quite differently. Consequently, decisions which produce system-wide effects made by distinct computers can often conflict, invariably causing inefficiency in resource management and therefore leading to downgraded performance.

To address these and related problems, a system is proposed which provides the following:
* a mechanism for monitoring events of interest in a distributed system;
* a mechanism for distributing monitored data throughout the distributed system;
* a mechanism which uses heuristic-based specifications to interpret received monitored data from remote sources so that appropriate actions can be taken when necessary.

The novelty and power of the proposed system lies in its application of expert system technology to deal with uncertain, incomplete, erroneous and out-of-date observation data which is inevitable when one tries to efficiently monitor remote events in a distributed system.




Download Full History