Recently, a paper co-authored by our Database Systems Research Group in collaboration with the University of Waterloo and the University of New South Wales was accepted by the international academic conference VLDB 2024. The first, second, and third authors of the paper are PhD students Jia-Min Hou, Zhan-Hao Zhao, and Master’s student Zhou-Yu Wang, respectively. The supervising professors are Xiao-Yong Du and Wei Lu. VLDB is a Class A international academic conference recommended by the China Computer Federation (CCF) and is one of the most prestigious conferences in the field of databases. The 2024 International Conference on Very Large Data Bases will be held in Guangzhou from August 25th to August 29th, 2024.
Title: AeonG: An Efficient Built-in Temporal Support in Graph Databases
Authors: Hou Jiamin, Zhao Zhanhao, Wang Zhouyu, Lu Wei, Jin Guodong, Dong Wen, Du Xiaoyong
Corresponding Author: Lu Wei
Abstract: Graphs in the real world are often dynamic and evolve over time. Storing the evolution of graphs and enabling temporal queries on them is crucial. However, existing approaches either incur high storage overhead or lack efficient temporal query support. To address these limitations, this paper proposes AeonG, a graph database with built-in temporal support. This method extends the traditional property graph model by introducing temporal features, thereby defining a temporal graph model. Based on this temporal graph model, we redesign both the storage engine and the query engine. The storage engine is divided into current storage and historical storage components. The current storage manages the latest versions of graph objects, while the historical storage manages previous versions of graph objects. This separation reduces the performance impact of querying the latest versions of graph objects. Additionally, to minimize historical storage overhead, we propose an anchor+delta strategy. This strategy periodically creates full versions (anchors) of graph objects and maintains each change (delta) between adjacent anchors of the same object. To improve temporal query processing efficiency, we introduce an anchor-based version retrieval technique in the query engine, which skips unnecessary traversal of historical versions. We conducted extensive experiments on real-world and synthetic datasets. Results show that compared to state-of-the-art methods, AeonG reduces storage consumption by up to 5.73 times and temporal query latency by 2.57 times, while only causing a 9.74% performance degradation for primary non-temporal queries.
Hou Jiamin is a Ph.D. student in the School of Information at Renmin University of China, majoring in Computer Application Technology. His advisors are Professor Du Xiaoyong and Professor Lu Wei, and his primary research focus is on graph database systems.
Lu Wei is a professor at Renmin University of China and a Ph.D. advisor. He is also a member of the Database Special Committee of the China Computer Federation (CCF). In recent years, he has focused on research in the fields of database theory and distributed databases. He has published over 50 papers in renowned international conferences and journals, including SIGMOD, VLDB, ICDE, ATC, VLDB Journal, and TKDE.
