A survey of optimization methods for transactional database in multi-core era

The design of traditional DBMS assumes magnetic disk as the storage device. Its optimization techniques are focused on the reduction of I/O cost. However, the database platforms in the future will be dominated by multi-core processors, large main memory and low-latency semiconductor storage, such as SSD. On such platforms, the entire data set can normally fit into main memory or high-speed storage. Thus, the performance bottleneck of query execution has shifted from disk I/O to CPU. The components for locking, latching, logging and buffer management of traditional DBMS were not originally designed for multi-core processors. These components severely prohibit the scalability of DBMS in multi-core architectures. Adaptation of traditional DBMS to new hardware is a common and necessary practice. This paper provides a survey of the recent optimization techniques proposed for DBMS on multi-core platforms. Meanwhile, the efforts by Renmin University of China on DBMS optimization on multi-core platform are introduced.