Embedded environment needs fast, credible and efficient mechanism of allocating memory. But existing allocating mode of OS is too complicated to be applied. For it's powerful performance, RTDBS expects to treat main memory database (MMDB) as rock-bottom support. So I/O for secondary storage will not be bottleneck any more. The goal of designing algorithm will be transferred to efficiently employing CPU and main memory. In tranditional disk database, index can't satisfy powerful performance in both storaging and accessing and employing main memory. Aiming at above problem, a mode of data-storage manager in main memory(MMDSM)must be studied. The mode has character of modern application. It is looked as the basic project of embeded active real-time database system.Based on a embed active real-time database system (ARTs-EDB) which being studied by ourselves, design and implementation of MMDSM are explored deeply. It reflects character of accessing fast in main memory.With respect to the characters of ARTs-EDB and beginned with character and function of MMDSM, the architecture of MMDSM in ARTs-EDB are introduced and it contains the main modules in MMDSM and interfaces to other subsystems.Using the memory-managing mechanism of OS for reference, this paper proposed a DB system-memory-manager mode, which has the features of modern applications. In terms of the definitions of system memory management of ARTs-DB, the object and realization strategy of ARTs-EDB's system memory manager is discussed, also, a realization method is introduced in detail.Our research breaks concept of designing traditional disk DBs, and refers to character of accessing fast in main memory. Thus, new physical structure and data format in MMDB are advanced based on existing partition-segment model. It is described in detail that data in MMDB is being storaged and accessed and segments are being exchanged between MMDB and Secondary Database.In this paper, configuration and character of balanced binary tree (AVL tree) and SB-tree (Sequential Binary tree) are analyzed. Realization and maintenance of SB-tree are depicted concretely. At last, a method of realizing multi-dimension index is discussed based on SB-tree. |