The Consistency Analysis of Secondary Index on Distributed Ordered Tables

The demand for multi-dimensional range query over Distributed Ordered Table (DOT) has become increasingly popular, however, the DOT does not support queries very well other than the primary key. One solution to this problem is indexing. Many indexing techniques are focusing on how to improve the query ability, but do not care about the consistency between the index table and base data table. This paper focuses on the relationship between the consistency and the performance about the indexing techniques. In this paper, we give the definition about the consistency between the index table and the base data table, and present the inconsistency window to measure the degree of consistency. We classify the existing indexing techniques into four categories: sync, async-compact, async-insert and async-simple, according to their implementation in terms of how the index is maintained. Experimental results confirm that async indexing technique have better performance compared with sync indexing technique but with worse consistency. For async indexing techniques, the performance of async-simple is about 1.86x to 6.26x compared with async-insert but with similar consistency.