Authors: Chunbo Lin, Jingdong Li, Xiaoling Wang, Xingjian Lu and Ji Zhang
Abstract: Window functions, despite being supported by all major database systems, are unable to keep up with the steeply growing size of data. Recently, some approximate query process (AQP) systems are proposed to deal with large and complex data in relational databases, which offer us a flexible trade-off between accuracy and efficiency. At the same time, Machine Learning has been adopted extensively to optimize databases due to its powerful ability in dealing with data. However, there have been few publications that consider using AQP techniques especially model-based methods to accelerate window functions processing. This work presents WFApprox, an AQP system based on Machine Learning models aims at efficiently providing an approximate answer for window functions. WFApprox uses Machine Learning models instead of massive data for query answering. Our experimental evaluation shows that WFApprox significantly outperforms the mainstream database systems over TPC-H benchmark.
