CliffGuard: A Principled Framework for Finding Robust Database Designs

摘要: A fundamental problem in database systems is choosing the best physical design, i.e., a small set of auxiliary structures that enable fastest execution future queries. Almost all commercial databases come with designer tools create number indices or materialized views (together comprising design) they exploit during query processing. Existing designers are what we call nominal; is, assume their input parameters precisely known and equal to some nominal values. For instance, since workload often not priori, it common for these optimize past workloads hopes queries data will be similar. In practice, however, noisy missing. Since do take influence such uncertainties into account, find designs sub-optimal remarkably brittle. Often, as soon deviates from past, overall performance falls off cliff, leading customer discontent expensive redesigns. Thus, propose new type robust against parameter uncertainties, so degrades more gracefully when deviate past. Users express risk tolerance by deciding on how much optimality willing trade attaining desired level robustness uncertain situations. To our knowledge, this paper first adopt recent breakthroughs theory optimization build practical framework solving most problems databases, replacing today's brittle principled world can guarantee predictable consistent performance.