RDS: a primitive for the maintenance of replicated data objects

A reliable data server (RDS) is a distributed system that encapsulates replicated data objects (either main memory or secondary storage copies on different sites) and their maintenance primitives. RDS is seen by its clients as a single entity regardless of the number of sites that constitute it. Since different applications may have different reliability requirements, RDS can provide uninterrupted service while tolerating a user-defined number of failures. To achieve a desired degree of resiliency free sites may be dynamically incorporated into RDS to bold additional copies of data objects. Under this approach, a site failure does not cause the abortion of transactions in progress and the availability of the replicated data is not limited during recovery. The design of the system is presented and some preliminary results of the evaluation of its performance are discussed