Broadcast and Epidemic Communication for Managing ReplicatedData

点击翻译按钮获取中文摘要

Many commercial database products and application tools are beginning to support data replication to provide ready access to data in distributed, mobile, and disconnected environments. Most of these systems have chosen to use asynchronous replication schemes in which updates to replicas are not synchronized. Although these schemes perform reasonably well for single object updates (i.e., environments such as file-systems), they are not suitable when a single update involves multiple objects (i.e., transactional environments). This project investigates the benefits of using synchronous and asynchronous broadcast mechanisms for managing replicated databases. Broadcast primitives, viz., reliable broadcast, causal broadcast, and atomic broadcast, will be used to design new protocols for synchronous replication. In synchronous replication, transactions accessing replicated data are synchronized with other transactions in the system. In asynchronous replication, transactions execute locally and other means are used to disseminate their effects globally as well as to detect and correct inconsistencies. The epidemic model of communication will be used as the transport mechanism here. Pessimistic as well as optimistic replication protocols will be developed to maintain replicas asynchronously. Finally, a simulation framework will be developed to evaluate the protocols for synchronous and asynchronous replication.