| The traditional database indexing uses a fixed reading-writing strategy and a fixed data structure. This will not produce any problems in traditional database, because it is based on disk, which is balanced in reading and writing operations. It is not necessary to differentiate the operations of reading and writing according to the performance. However, NAND Flash is an unbalanced device, whose writing operation costs much higher than reading, furthermore, which has huge difference in accessing the different packages of NAND Flash. According to the reasons above, the traditional database is not suitable for NAND. To make the indexing work efficiently and commonly on variable devices, a new flexibly dynamic indexing mechanism is becoming desiderative.The work of this thesis is to make use of the characteristic of NAND Flash to improve the performance of database indexing mechanism. After summarizing the storage system based on NAND, we describe a design and implementing of dynamic indexing mechanism based on B+ tree. The mechanism integrates the strength of disk mode and log mode, determines and transfers to the best mode of each node of B+ tree when runtime, and makes it performance better on variable devices.The design goals of dynamic indexing are commonality and high efficiency. For commonality, the mechanism determines the more suitable mode for run time environment and translates the data mode when necessary. For high efficiency, it does optimization for every node of B+ tree and so performances better than any other fixed mechanism.This thesis discusses the design and implementation of the dynamic indexing mechanism, and analyzes every module, including the node transferring table and log buffer in BFTL and the mode transferring algorithm in B+ tree management. Besides, the thesis also gives the relationship between the size of B+ tree node and the performance of indexing. Finally, experiment shows that this dynamic indexing mechanism is more efficient than disk and log mode. |
PDF Full Text Request