Fault-Tolerant Algorithms for Distributed Databases

分布式数据库中数据的可用性可以通过复制来提高，其中一个数据对象可能有多个驻留在不同站点上的副本。 如果数据被复制，则在发生故障后它可能仍然可用。 然而，数据可用性的增加可能会导致操作执行成本的增加。 本文提出的研究旨在开发以低成本提高数据可用性的方法。 此外，我们探索的机制允许数据在站点或通信故障的情况下保持可访问性，甚至当此类故障导致网络分区时也是如此。 我们建议开发一种新的容错并发控制协议范例。 这种范式加深了我们对此类协议的理解，并为我们开发新协议提供了新的见解。 我们特别提出了开发协议，允许执行只读事务以及在具有明确定义的信息流模式的数据库中高效执行事务。 最后，我们建议研究提高抽象数据对象可用性的方法，其中有关数据库中数据对象的语义信息可用。

The availability of data in a distributed database can be increased by replication, where a data object may have several copies residing on different sites. If data is replicated, it may still be available after failures. However, this increase in data availability may result in an increase in the cost of opertion execution. The research proposed herein seeks to develop methods that increase data availability at low costs. Futhermore, the mechanisms we explore allow the data to remain accessible in spite of site or communication failures, even when such failures lead to network partitioning. We propose to develop a new paradigm for fault tolerant concurrency control protocols. This paradigm furthers our understanding of such protocols, and provides us with new insights for the development of new protocols. We specifically proposed development protocols that allow the execution of read-only transactions and the efficient execution of transactions in databases with well defined information flow patterns. Finally, we propose to investigate methods for increasing the availability of abstract data objects where semantic information is available about the data objects in the databaase.