持久对象框架中基于多级访问模式的对象预取技术研究与实现

Research and Implementation of Multi-level Access Pattern-based Prefetching in Persistent Object Framework

在基于关系数据库和对象关系映射的持久对象框架中,对象之间通常通过对象引用和各种集合属性将对象相互关联起来,组合成更为复杂的复合对象。应用程序对这些复合对象的访问则是通过使用这些属性逐个访问成员对象来完成。这种在多个成员对象之间的导航操作导致了客户端和后端数据库系统之间的获取操作大幅度增加,从而导致严重的性能问题。对象预取技术根据某种策略,将应用程序可能访问到的对象成组或批量地预先从数据库中装载到客户端,从而减少了应用程序对后端数据库系统进行查询的次数。本文对现有各种对象预取技术并对其进行分析分类,在此基础上,提出了一种基于多级访问模式的对象预取技术。最后,介绍了该算法在软件构件平台StarC-CM的持久对象框架中的实现。

In the persistent object framework based on RDBMS and O/R mapping, objects connect each other via references and collection attributes, and applications navigate from one to another in these objects. These navigations dramatically increased the fetch operations between client and back-end database server, and brought serious performance problem. Object prefetching techniques load objects that the application maybe access successively from database to client-side cache by group or in batch based on some kind of predication policy. This paper classifies various object prefetching policy, and promots a new kind of prefecthing algorithm based on multi-level access pattern. Furthermore, a prototype implementation of this algorithm in the persistent object framework designed for our software component platform StarCCM.