Managing risk and uncertainty in query optimization

查询优化器作为数据库系统中最重要、最复杂的组件之一，是管理大型数据库的关键，其主要任务是选择最佳的查询执行方式。通常有几种不同的方法(执行计划)来执行给定的查询，它们可能在使用的索引、执行的操作符的顺序等方面有所不同，但都会产生相同的正确结果。查询优化器生成并检查候选执行计划，并决定遵循哪个计划执行。优化器的质量非常重要，因为计划的成本可能会有很大的差异，选择一个糟糕的计划可能会导致性能极差。现有的大多数查询优化方法的目标是选择估计成本最低的执行计划。然而，在许多情况下，这是不够的，因为其他素质，如稳健性和可预测性，也可以同样重要。像许多现实生活中的决策问题一样，查询优化本身就包含风险和不确定性，在选择计划时必须考虑这些风险和不确定性。我们研究的首要目标是开发能够有效利用数据库系统中各种可用信息源的系统、方法和技术，以降低查询优化决策的风险和不确定性，使查询处理更加健壮和可预测。

The query optimizer, as one of the most important and most complex database system components, is key to the management of large databases.  Its major task, query optimization, is to choose the best way to execute a query. Often there are several different ways (execution plans) to perform a given query, which might differ in the indices used, the orders of operators executed, etc., but all lead to the same correct result. The query optimizer generates and examines the candidate execution plans, and makes the decision as to which plan to follow for execution. The quality of the optimizer is very important, because costs of plans can vary dramatically and choosing a bad plan might result in extremely poor performance.Most existing approaches to query optimization aim at choosing the execution plan with the least estimated cost. However, in many cases, this is insufficient, as other qualities, such as robustness and predictability, can be equally important. Like many real-life decision making problems, query optimization inherently involves risks and uncertainties, which must be taken into consideration when choosing a plan.  The primary goal of our research is to develop systems, methods, and techniques that can effectively utilize various sources of information available in a database system, in order to reduce the risk and uncertainty of query optimization decisions, and make query processing more robust and predictable.