Adaptive Dataflow: Eddies, SteMs and FLuX

这个项目提出了一个议程，详细研究为查询处理中的自适应而设计的三个关键操作符：涡流用于在面对不断变化的数据统计和系统行为时，对查询计划中的操作符进行连续的、适应性的重新排序。词干将查询处理中固有的状态管理暴露为物理代数中的第一类运算符，从而允许查询处理备选方案之间的自适应竞争和杂交。FLuXen提供跨并行机的自适应负载平衡和容错。利用电报系统基础设施，该提案的议程是在传统和新的查询处理挑战的背景下，研究高效、健壮的涡流、茎和通量的机制和政策。最终目标是提供一种并行查询处理体系结构，它可以在完美的环境中与传统体系结构运行得几乎一样好，更重要的是，当面对大规模环境中常见的各种缺陷和变化时，可以自动适应高效和健壮的运行。

This project sets out an agenda to study in detail three key operators designed for adaptivity in query processing: Eddies serve to do continuous, adaptive reordering of operators in query plans in the face of changing data statistics and system behavior. SteMs expose the state management inherent in query processing as a first class operator in a physical algebra, allowing for adaptive competition and hybridization among query processing alternatives. FLuXen provide adaptive load balancing and fault tolerance across parallel machines. Using the Telegraph system infrastructure, this proposal's agenda is to investigate mechanisms and policies for efficient, robust eddies, SteMs and FluXen in the context of traditional and new query processing challenges. The end goal is to provide a parallel query processing architecture that can run nearly as well in perfect circumstances as a traditiona architecture, and -- more importantly -- that can automatically adapt to run efficiently and robustly when faced with the kinds of imperfections and variations commonly seen in large-scale environments.