Enforcing crash failure semantics in distributed systems with fine-grained object mobility

Migration is a powerful technique in distributed systems providing many benefits. The granularity of migration ranges from the coarse-grained movement of whole processes to the fine-grained mobility of individual objects which provides more flexibility and control. One of the costs of fine-grained mobility is an increase in the complexity of programming with respect to failures. Classic fault-tolerance techniques for distributed systems cannot be applied in systems with fine-grained object mobility due to the unacceptable overhead of applying these techniques to many small objects. We discuss a group service that allows programmers to apply classic distributed system fault-tolerance techniques to systems with fine-grained object mobility. This service enforces the condition that all objects in a group are either all available or all failed, and has been implemented in the Emerald language and runtime environment. Examples using the group service include a fault-tolerant name server and a fault-tolerant distributed system monitor.Science, Faculty ofComputer Science, Department ofGraduat