A dynamic accessibility protocol for replicated databases

摘要

In this paper, we present a protocol for managing replicated data in a distributed system. The protocol tolerates both site and communication failures even when they lead to partitioning. Two of the main features of the protocol are the availability levels it provides, and the low communication costs required of user transactions to attain these levels. In particular, an object may be accessible even after the failure of most of the sites where copies of the objects reside. Furthermore, an object can be read by accessing a single copy, and in general, the cost of executing operations is less than that required by protocols that attain similar levels of availability. Finally, the protocol provides the database designer with the flexibility of determining both the degree of availability as well as the cost of executing the different operations of a data object.