Research Of Storage Management In Main Memory Database

Because of that the memory can stored all or most of the data of Main MemoryDatabase (MMDB), there is little I/O during transaction processing of MMDB. So, whichis different from traditional Disk-Resident Database (DRDB), reducing I/O is no more thegoal of optimization of MMDB. The different means that the various methods for DRDBare no longer suitable for MMDB, it is necessary to propose new methods for processingdata in MMDB based on the features of itself. As the foundation of MMDB, storagemanagement is very important. Storage management influences not only the performancesignificantly, but also the high-level functions of MMDB. So it is very meaningful tostudy the storage management of MMDB.Based-on the above requirements, this thesis studies the storage management in MMDB,and presents the design and implementation of the corresponding storage manager.Specifically, the study includes data organization, memory management, and indexmechanism. The contributions of the thesis mainly include: presents the storage hierarchyof the storage manager and the concrete organization of records based-on the analysis ofstorage hierarchy in MMDB; improves the performance of the storage manager byimproving the format of records; designs and implements a memory management methodby using multiple memory pools, the method can efficiently utilizes the memory; designs aflexible startup load policy and a data exchange policy; deeply analyze the features andusability of the hash index, T-tree index and cache-conscious index mechanism, andpresents the details of design, implementation and optimization methods of T-tree index.At the last, the experiments show that the size of index node has influence onperformance of index, and the improvement of record format significantly improve theperformance of the storage manager. And the experiment contrasted to FastDB also showsthat the performance of the storage manager is equivalent to or higher than theperformance of FastDB, which means that performance of the storage manager is feasiblein practice.