CETLH: A NEW HISTOGRAM-BASED CARDINALITY ESTIMATE APPROACH

For the mainstream relational database management systems, histograms play important roles in cardinality estimate. The main histogram-based cardinality estimate approaches can be classified into two categories: the proactive approach and the reactive approach. For the former, histograms are constructed and updated by periodical data scans. Data updates can not be incorporated into a histogram in real time, so between two data scans, large errors of cardinality estimate will occur. For the latter, data scans are avoided, as an alternative, query feedback records (QFRs) are collected to construct and update histograms. Although data updates can be incorporated into a histogram by replacing stale QFRs in real time, the cost of time is very expensive. For each histogram reconstruction, all buckets in the histogram will be recalculated and the large amount of computation leads to the inefficiency of the reactive approach. In this paper, we propose a novel QFR-based cardinality estimate approach which can balance the efficiency issue and the data update issue: on the one hand, it can improve the efficiency of QFR-based cardinality estimate to a practical level; on the other hand, it can incorporate data updates into a histogram in real time to fully ensure the accuracy of cardinality estimate. In our approach, a serial of small second-level histograms covering different parts of the whole value range of an attribute will be constructed. These second-level histograms can be updated independently over time to ensure the performance of the approach. As the update of each second-level histogram, QFRs still play their important roles in incorporating data updates into the histogram in real time. Extensive comparison experiments fully demonstrate the advantages of our approach in performance and accuracy.