Recursive Computation Over Relational Data (RECORD)

关系数据的递归计算 (RECORD)

• 批准号: 511062611
• 负责人: Professor Dr. Torsten Grust
• 金额: --
• 依托单位: Lehrstuhl für Datenbanksysteme
• 依托单位国家: 德国
• 项目类别: Research Grants
• 资助国家: 德国
• 项目状态: 未结题
• 来源: https://gepris.dfg.de/gepris/projekt/511062611?language=en
• 关键词: Recursive; Computation; Over; Relational; Data

Move your computation close to the data! This age-old mantra of the database community asserts that we can expect a SQL query engine with immediate access to the data to perform significantly better than an external processor to which we have to ship the data first. The lore holds up if the computation is query-like and primarily involves filtering, data (re-)combination, grouping, or aggregation. It is much less clear how complex algorithms that rely on arbitrary iterative control flow or recursion can be efficiently evaluated inside the same SQL engines. With the advent of SQL:1999, contemporary database engines started to support forms of recursion. The associated language constructs, however, exhibit syntactic restrictions, are based on semantics that are tough to grasp for regular developers, or exhibit sobering runtime performance that often render iterative or recursive SQL impractical. Project RECORD explores compilation and implementation techniques that (1) admit the formulation of iterative and recursive algorithms in a readable, concise (even elegant) fashion and (2) use relational database systems as efficient and scalable runtime environments that perform the computation right next to the data. We adopt established techniques originally developed by the (functional) programming language community, then adapt and bend these ideas so that they apply to recursive SQL functions as well as iterative PL/SQL procedures written in an imperative style. Our focus is on non-invasive approaches that do not turn existing database technology on its head: we thus map functions and procedures to the native, plain SQL recursion constructs already built into off-the-shelf database systems. We take the freedom, however, to apply surgical changes to database kernels where we anticipate that the runtime performance or the systems' space usage can benefit. There is no shortage of data-intensive problem domains whose need for such in-database computation only ever goes up. RECORD will study the core data structures and algorithms of these domains to test-drive its results and to prove that recursive computation over relational data can indeed be practical and efficient.

让你的计算靠近数据！这个数据库社区的古老咒语断言，我们可以期望能够立即访问数据的SQL查询引擎比我们必须首先将数据发送到的外部处理器执行得好得多。如果计算是类似查询的，并且主要涉及过滤、数据（重新）组合、分组或聚合，则该规则适用。依赖于任意迭代控制流或递归的复杂算法如何在相同的SQL引擎中有效地求值则不太清楚。随着SQL:1999的出现，现代数据库引擎开始支持递归形式。然而，相关的语言结构表现出语法限制，它们基于对普通开发人员来说很难掌握的语义，或者表现出令人担忧的运行时性能，这通常会使迭代或递归SQL变得不切实际。RECORD项目探索的编译和实现技术(1)允许以可读、简洁（甚至优雅）的方式制定迭代和递归算法，(2)使用关系数据库系统作为高效和可扩展的运行时环境，在数据旁边执行计算。我们采用最初由（函数式）编程语言社区开发的成熟技术，然后对这些思想进行调整和弯曲，使其适用于递归SQL函数以及以命令式风格编写的迭代PL/SQL过程。我们的重点是非侵入性的方法，这些方法不会彻底改变现有的数据库技术：因此，我们将函数和过程映射到已经内置到现成数据库系统中的本机、普通SQL递归构造。但是，我们可以自由地对数据库内核应用外科手术式的更改，因为我们预计运行时性能或系统的空间使用可以从中受益。对这种数据库内计算的需求不断增加的数据密集型问题领域并不缺乏。RECORD将研究这些领域的核心数据结构和算法，以测试其结果，并证明在关系数据上的递归计算确实是实用和高效的。