摘要： 通过对关系数据库管理系统PostgreSQL 的大对象数据缓冲机制的分析，提出了基于流数据的大对象数据缓冲机制。该机制针对传统大对象数据缓冲所使用的页存储机制存在的性能缺陷，以流数据模型来取代一般的块页读取机制，并根据流数据的存储特点构建基于流数据管道的大对象数据缓冲器结构。该结构在内存中维护一个远小于大对象数据规模的可常驻内存的概要数据结构，从而可以迅速得到用户想要的数据，避免了冗余的磁盘I/O。

关键词: 大对象；缓冲机制；流数据；概要数据结构

Abstract: This paper creates a new buffer strategy based on stream data with the analysis of PostgreSQL. This strategy replaces the traditionalpage accessing model with stream model to get better performance. In addition, under this new buffer strategy, it recreates the buffer structure forLOB using stream pipes. This buffer structure maintains a synopsis data structure on memory which has less size of data than LOB. Through thissynopsis data structure, it can get needed data more quickly with avoiding redundant disk I/O.

Key words: Large object (LOB); Buffer strategy; Stream data; Synopsis data structure