By reducing I/ O costs, index push-downs eliminate fact table rows more efficiently than bit-vector push-downs, but their processing cost is higher. 通过减少I/O成本,索引下推可以比位向量下推更有效地消除事实表行,但处理成本比较高。
For multiple index push-downs, multi-index scan is used to combine RIDs from different index scans, then skip-scan on the fact table is performed. 对于多索引下推,多索引扫描用于组合来自不同索引扫描的RID,然后就可以执行对事实表的跳跃式扫描了。
The optimizer chooses the best number of push-downs based on cost estimation, such that the work required to perform push-down does not outweigh the benefit for applying the push-down filter. 优化器根据成本估算选出最佳的下推次数,以便执行下推所需的工作量不会超出应用下推过滤的好处。
Multiple-bit vectors can be used for multiple bit-vector push-downs. 多位向量可用于多位向量下推。
Index push-downs require only a single-column index on the fact table, although they can use a multi-column index if the required foreign-key column is the first key in the index. 索引下推只需要事实表的单列索引,虽然在所需的外键列是索引的第一个键的情况下可以使用多列索引。