• Journal of the Korea Society of Computer and Information
• /
• v.16 no.8
• /
• pp.29-38
• /
• 2011

Unlike in traditional row-oriented database systems, a column-oriented database system stores data in column-oriented and not row-oriented order. Recently, research results revealed the effectiveness of column-oriented databases for applications such as data warehouse and decision support systems that access large volumes of data in a read only manner. In this paper, we investigate the join strategies for column-oriented databases and prove the effectiveness of column-oriented databases in data warehouse systems. For unbiased comparison, the two database systems are analyzed using the star schema benchmark and the performance analysis of a star schema join query is carried out. We experimented with well-known join algorithms and considered early materialization and late materialization join strategies for column-oriented databases. The performance results confirm that star schema join queries perform better in terms of disk I/O cost in column-oriented databases than in row-oriented databases. In addition, the late materialization strategy showed more performance gain than the early materialization strategy in column-oriented databases.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2011.04a
• /
• pp.1214-1217
• /
• 2011

데이터 레코드를 가로(row-wise)로 저장하는 기존의 데이터베이스를 Row-지향 데이터베이스, 세로(column-wise)로 저장하는 데이터베이스를 Column-지향 데이터베이스라 정의하자. 본 논문에서는 Row-지향 데이터베이스와 Column-지향 데이터베이스에서 분석 workload 형태의 조인 질의를 처리하여 비교 우위 성능을 보이는 데이터베이스 시스템을 고찰하고자 한다. 객관적인 성능 실험을 위해 분석적 모델인 스타 스키마 벤치마크를 이용하였다. Nested Loop 조인과 Sort Merge 조인 기법을 사용한 실험에서 Column-지향 데이터베이스의 성능이 우수하게 나타났음을 확인할 수 있다.