Managing replicated copies in very large distributed systems

Data management in very large distributed systems (VLDS) is considered. A replication strategy that can offer a tradeoff between cost and quality of service is discussed. Imperfect replicated data, called quasi-copies, are considered in conjunction with the notion of file stashing. Stashing a file means placing copies of that file in several nodes of a computer network to augment the availability of the file, i.e. to be able to access the file when the file server that holds it is no longer reachable. The stash propagation problem of how to efficiently propagate updates to the stashed copies throughout the system is studied