摘要： 根据对象分布相对稳定的特点，选择与固定栅格对应的、代表对象分布情况的部分叶子节点作为容纳新记录的种子节点，新记录可直接与种子节点合并而无须遍历R树。随机选择部分无法合并的记录作为种子记录，对活动记录进行简单有效的分组，以插入种子记录的代价实现批量插入。上述2种方法考虑了R树的空间聚簇特性，可在一次更新中完成多项插入与删除，减少了对节点的写操作及对R树的遍历次数。实验证明，该机制在降低索引维护I/O开销的同时保证了查询效率。

关键词: R树维护, 栅格, 批量插入

Abstract: According to the relative stability of objects’ distribution and based on fixed grids, some leaf nodes representing the distribution of objects are selected as seed-nodes for new records to directly merge new records into seed-nodes without traversing R-tree. Some records which can not be merged into seed-nodes are chosen randomly as seed-records. Such records are grouped simply but effectively, so that they can be inserted into R-tree with the cost of inserting seed-nodes. The strategies pay attention to R-tree’s spatial clustering, and perform multiple inserting and deleting in one write-operation, decreasing the demand of writing and traversing R-tree. Experiments show that the strategies reduce the I/O cost of R-tree maintaining without affecting the query performance.

Key words: R-tree maintaining, grid, batch-inserting

中图分类号: 

• TP311.12