Dynamic modeling of replicated objects for dependable soft real-time distributed object systems

Abstract

Soft real time distributed object systems that must be dependable typically replicate application objects on different processors within the distributed system. The paper presents an algorithm for determining the degree of replication of each application object, that aims to maximize the system utility while respecting the resource constraints. The system utility captures the relative importance of the individual objects for the application tasks and the relative rewards of executing the application tasks. The algorithm determines an initial replication degree for each application object and adjusts the replication degree dynamically at run time.