benchmark

A Benchmark For Online Index Selection#1 #2Karl Schnaitter , Neoklis Polyzotis#Computer Science Department, Univ. of California Santa CruzSanta Cruz, CA, USA1karlsch@soe.ucsc.edu2alkis@soe.ucsc.eduAbstract—Online approaches to physical design tuning have experimental methodology for benchmarking the performancereceived considerable attention in the recent literature, with a of online tuning algorithms.focus on the problem of online index selection. However, it is In this paper, we make concrete contributions in the pre difficulttodrawconclusionsontherelativemeritsoftheproposedviously outlined directions. First, we propose a principledtechniques, as they have been evaluated in isolation usingbenchmarking methodology for evaluating the performance ofdifferent methodologies. In this paper, we make two concretecontributionstoaddressthisissue.First,weproposeabenchmark online tuning algorithms. The proposed benchmark consists offor evaluating the performance of an online tuning algorithm in several workload suites, designed to exercise specific aspectsa principled fashion. Second, using the benchmark, we present of a tuning algorithm. The workloads are constructed using aa comparison of two representative online tuning algorithmsgeneral methodology that can be used to generate additionalthat are implemented in the same database system. The resultsinteresting suites. Using the benchmark, we then presentprovide interesting insights on the behavior of theseand ...