Abstract: Indexes play a pivotal role in improving the performance of queries. One of the important tasks of database administrators is to select appropriate indexes for databases. However, with the increasing complexity of database workloads, the rising amount of data volumes, and the gradually increasing number of tables, it has become more and more difficult to artificially analyze workload cost, select appropriate indexes, and estimate database spatial changes. This paper designs a load-oriented automatic indexing recommendation tool: CedarAdvisor. It collects workload automatically from logs and counts query frequency. First, it generates a candidate index for each query and then evaluates the respective revenue and cost of these candidates. Finally, it generates an indexing recommendation for the complete workload through a dynamic programming algorithm. We verify the effectiveness of the tool through experiments on the distributed database.

Key words: database, workload, index recommendation