SQL Server查询优化器的测试：挑战性，技术性和经验性外文翻译.doc

外文翻译 原文 Testing SQL Server’s Query Optimizer: Challenges, Techniques and Experiences Material Source:IEEE Data Engineering Bulletin Author: Leo Giakoumakis, Cesar Galindo-Legaria 1 Introduction Today’s query optimizers provide highly sophisticated functionality that is designed to serve a large variety of workloads, data sizes and usage patterns. They are the result of many years of research and development, which has come at the cost of increased engineering complexity, specifically in validating correctness and measuring quality. There are several unique characteristics that make query optimizers exceptionally complex systems to validate, more so than most other software systems. Query optimizers handle a practically infinite input space of declarative data queries (e.g. SQL,XQuery), logical/physical schema and data. A simple enumeration of all possible input combinations is unfeasible and it is hard to predict or extrapolate expected behavior by grouping similar elements of the input space into equivalence classes. The query optimization process itself is of high algorithmic complexity, and relies on inexact cost estimation models. Moreover, query optimizers ought to satisfy workloads and usage scenarios with a variety of different requirements and expectations, e.g. to optimize for throughout or for response time. Over time, the number of existing customers that need to be supported increases, a fact that introduces constraints in advancing query optimization technology without disturbing existing customer expectations. While new optimizations may improve query performance by orders of magnitude for some workloads, the same optimizations may cause performance regressions (or unnecessary overhead) to other workloads. For those reasons, a large part of the validation process of the query optimizer is meant to provide an understanding of the different tradeoffs and design choices in respect to their impact across different customer scenarios. At the same time, the