Precise, compact, and fast data access counters for automated physical database design

Today’s database management systems offer numerous tuning knobs that allow an adaptation of database system behavior to specific customer needs, e. g., maximal throughput or minimal memory consumption. Because manual tuning by database experts is complicated and expensive, academia and industry devised tools that automate physical database tuning. The effectiveness of such advisor tools strongly depends on the availability of accurate statistics about the executed database workload. For advisor tools to run online, workload execution statistics must also be collected with low runtime and memory overhead. However, to the best of our knowledge, no approach collects precise, compact, and fast workload execution statistics for a physical database design tool. In this paper, we present data structures that solve the problem of providing workload execution statistics with high precision, low memory consumption, and low runtime overhead. In particular, we show how existing approaches can be combined and for which advisor tools, new data structures need to be designed. We evaluate our data structures in a prototype of a commercial database and show that they outperform previous approaches using real-world and synthetic benchmarks