一种面向分布式数据库的多并发OLAP型查询性能预测方法

A concurrent OLAP query performance prediction approach for distributed database

摘要:

数据库查询性能预测即预测查询的执行延迟时间。数据库查询延迟预测技术能够用于查询调度和查询进度显示等。当前的查询性能预测技术都具有一定的局限性，不能准确预测分布式数据库中多并发查询导致CPU、I/O资源争用情形下的查询延迟。针对这一问题，该论文提出了一种在分布式数据库中预测并发OLAP型查询延迟的技术，该技术通过建立查询干扰度和查询敏感度模型来衡量资源（I/O，网络）竞争的激烈程度，并据此来预测不同执行环境下的查询性能。TPC-DS的实验结果表明，分析型查询预测的误差率在25%以下，说明该技术能够较准确地预测查询执行时间。

Abstract:

Database query performance prediction is predicting query latency time. Database query latency prediction technology can be used for query scheduling and query progress indicator. However, current query performance prediction techniques have some limitations and often to be hard to accurately predict the query latency caused by concurrency queries in distributed database, because of CPU and I/O resource contention. In order to solve this problem, this paper proposes a technique to predict concurrent OLAP query latency in distributed database. The proposed method measures the intensity of competition in resource (I/O, network) by setting up CQI(Concurrent Query Interference) and QS(Query Sensitivity) models , And predicts the query performance under different execution environments accordingly. The experimental results of TPC-DS show that the mean relative error of analytical query prediction is less than 25%, which shows that the technology can predict the query execution time more accurately.