并发内存OLAP查询优化技术研究

发布时间：2018-04-18 15:06

本文选题：并发OLAP查询处理 + 数组索引引用　；参考：《计算机研究与发展》2016年12期

【摘要】：基于多核处理器硬件技术和高并发查询负载需求,近年来的研究不仅关注于一次一查询模式的查询优化技术,而且也关注于一次一组模式的查询优化技术.通过将并发查询转换为共享负载,一些低访问延迟的操作,如磁盘I?O、cache访问,可以被多个并发的查询所共享.当前的研究通常基于共享查询操作符,如扫描、连接、谓词处理等,通过生成全局执行计划优化并发查询.对于复杂的分析型负载,如何创建优化的执行计划是一个具有挑战性的问题.在广泛使用的星形模型的基础上提出一种模板OLAP查询执行计划来简化查询执行计划,以达到最大化查询操作符利用率的目标.1)提出了基于代理键的连接索引技术,将传统的基于值探测的连接操作转化为内存数组索引引用(AIR),使连接操作的CPU效率更高并且支持聚集计算的后物化;2)并发查询的谓词处理简化为cache line敏感的谓词向量,在单次cache line访问中最大化并发查询谓词计算性能;3)通过多核并行实现技术在SSB基准上进行测试.实验结果表明:共享扫描和共享谓词处理能够将并发OLAP查询处理性能提升1倍.
[Abstract]:Based on the hardware technology of multi-core processor and the requirement of high concurrent query load, the research in recent years has focused not only on query optimization technology in one query mode, but also on query optimization technology in one set of query mode at a time.By converting concurrent queries to shared loads, some low-access latency operations, such as disk Igno cache access, can be shared by multiple concurrent queries.The current research is usually based on shared query operators such as scan join predicate processing and optimize concurrent queries by generating a global execution plan.For complex analytical loads, how to create optimized execution plans is a challenging issue.Based on the widely used star model, a template OLAP query execution plan is proposed to simplify the query execution plan to maximize the query operator utilization.The traditional join operation based on value detection is transformed into a memory array index to refer to AIRI, which makes the CPU of join operation more efficient and the predicate processing of concurrent query which can support aggregate computation be simplified to cache line sensitive predicate vector.In single cache line access, the performance of concurrent query predicate computation is maximized. (3) Multi-core parallel implementation technique is used to test the SSB benchmark.Experimental results show that shared scanning and shared predicate processing can double the performance of concurrent OLAP query processing.
【作者单位】：数据工程与知识工程教育部重点实验室(中国人民大学);中国人民大学信息学院;中国调查与数据中心(中国人民大学);中国气象局国家卫星气象中心;
【基金】：国家“八六三”高技术研究发展计划基金项目(2015AA015307) 中国人民大学科学研究基金(中央高校基本科研业务费专项资金资助)项目(16XNLQ02)~~
【分类号】：TP311.13

【相似文献】