CubeFlow: Accelerating Multidimensional Query Chains for Interactive Data Analytics
Year: 2026
Type: conference-paper
Abstract: Interactive data analytics tools enable users to explore data incrementally, often generating a sequence of dependent SQL queries, called a SQL chain. Even for moderately sized datasets, executing such chains on traditional database systems incurs significant overhead due to multiple rounds of intermediate aggregations, joins, and other compute?intensive operations. This paper focuses on multidimensional query chains and presents CubeFlow, an acceleration layer that sits between the analytics platform and the underlying database. By decoupling dimension columns from measure values, CubeFlow represents intermediate results as cube structures and models execution as a sequence of cube transformations along the chain. We introduce shadow dimensions to preserve regular cube structures across transformations in the chain, and propose coordinate?based aggregation and join algorithms that reuse dimension mappings and hash integer coordinates rather than raw attribute values. We implement a single?threaded prototype of CubeFlow and evaluate it on six multidimensional query chains over TPC?H, Online Retail, and Stocks. Despite the overhead of data conversion and coordination, our approach achieves average speedups of 5.4× over DuckDB and 7.6× over ClickHouse under single?threaded settings.
Authors: Gang Qian, Qinghua Zhao, Zhengrui Qian, Hongru Chen
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7225091314