Incremental Materialized View Maintenance on Distributed Log-Structured Merge-Tree

Abstract

Modern database systems are in need of supporting highly scalable transactions of data updates and efficient queries over data simultaneously for the real-time applications. One solution to reach the demand is to implement query optimization techniques on the online transaction processing (OLTP) systems. The materialized view is considered as a panacea to improve query latency. However, it also involves a significant cost of maintenance which trades away transaction performance. In this paper, we develop materialized views on a distributed log-structured merge-tree (LSM-tree), which is a well-known structure adopted to improve data write performance. We examine the design space and conclude several design features for the implementation of view on LSM-tree. An asynchronous approach with two optimizations are proposed to decouple the view maintenance with transaction process. Under the asynchronous update, we also provide consistency query for views. Experiments on TPC-H benchmark show our method achieves better performance than straightforward methods on different workloads.