Decentralized Resource Management and Fault-Tolerance for Distributed CORBA Applications

Assigning an application’s fault-tolerance properties (e.g., replication style, checkpointing frequency) statically, and in an arbitrary manner, can lead to the application not achieving its target resilience and performance. The resource management infrastructure that we have developed transparently determines a CORBA application’s resource usage and its rate/pattern of invocation across a distributed system. Using this information, our infrastructure makes more informed decisions about the application’s fault-tolerance properties, and dynamically adapts these decisions, as faults occur, and as resources are added and removed from the system. We have designed our prototype to be decentralized so that it is scalable and does not itself constitute a single point of failure.