A data warehouse collects and integrates data from multiple, autonomous and heterogeneous sources. The data warehouse effectively maintains materialized views over the source data. There are read-only transactions and an update transaction in data warehousing environments. Clients perform complex read-only queries to get information for decision making support on the materialized views. The materialized views are refreshed periodically by an update transaction, which propagates large batch updates from base tables. In current warehousing systems, the update transaction usually is isolated from clients read activity, limiting availability of the data warehouse. In this paper, in order for an update transaction to run concurrently with read-only transactions, we propose a concurrency control scheme for upper-level transaction management in data warehouse systems, called version-stamp ordering, VO for short, based on the modified multi-version time-stamp ordering concurrency control algorithms.
The performances of VO scheme, OSL, and MVNL have been evaluated through a simulation approach under uniform workloads in order to probe their performance tradeoffs. The simulation results indicate that VO is capable of providing better performance across a wide range of workloads. The following essential characteristics of VO lead to its better performance: (1) it significantly reduces the number of versions, (2) it keeps simplicity of concurrency mechanism using version-stamp ordering, (3) it provides an enhanced level of concurrency.