Managing object groups in fault-tolerant distributed object systems

Real-time distributed object systems require dynamic configuration management to accommodate object groups with different availability and resource requirements. We have built a configuration management system that uses the application's real-time and availability requirements to determine the degree of replication for object groups and the placement of members on processors. The degree of replication is driven by the rewards of object groups, the probability of failure of groups, and the importance of groups to the system. Monitoring mechanisms measure the usage of resources and detect resource overloads and object member failures. Our experimental results show that the system determines the replication of groups efficiently using current resource measurements, and gracefully allocates object members to processors when processor overloads are detected.