An experimental study of semantics-based concurrency control protocols

摘要

The read/write model, which models a database by a collection of data objects that can only be read or written by transactions, has been the traditional underlying data access model for database applications. However, the limited “concurrency” provided by the read/write model may disable some applications to deliver a high performance even when the applications are run on advanced hardware platforms. To address this issue, some researchers have recently proposed to enhance concurrency by exploiting the semantics of data objects. In particular, an abstract data type model is used. The concurrent control algorithms derived from this model have been shown theoretically, to provide more concurrency as the underlying data types and operations offer a much richer semantics than that of the read/write model [10], [15], [28], [29], [31], [34]. However, the practical performance characteristics of these algorithms have not been studied extensively. Therefore, we are motivated to build a testbed to investigate the overhead and the performance of these protocols [29], [33], [34] in an attempt to understand the practical characteristics of these protocols and to define concurrency from a practical point of view.