REFERENCE ARCHITECTURE FOR DISTRIBUTED SYSTEMS MANAGEMENT

Management of computing systems is needed to ensure efficient use of resources and provide reliable and timely service to users. Distributed systems are much more difficult to manage because of their size and complexity, and they require a new approach. A reference architecture for distributed systems management is proposed that integrates system monitoring, information management, and system modeling techniques. Three classes of system management-network services and devices, operating system services and resources, and user applications-are defined within this framework, and a detailed hospital application is presented to clarify the requirements for managing applications. It is argued that the performance management of distributed applications must be considered from all three perspectives. Several management prototypes under study within the COnsortium for Research on Distributed Systems (CORDS) are described to illustrate how such an architecture could be realized.