Research And Design On The Key Technology Of Memory Database

Memory Database(MMDB) is rapidly developing in the recent years. Due to the development of internet and big data, the efficiency of database becomes more significant, where the traditional disk database cannot handle. The development of hardware technology reduces the cost of the in-memory manufacture, which provides the feasibility of Memory Database. Currently, there are many No SQL databases developed that are based on in-memory technology. The advantage of No SQL database is at its fast reading\writing ability and the strong scalability. These advantages make the No SQL database suitable for processing large amount of data and used in distributed system. On the other side, there are some defects in the No SQL database yet. First, it reduces the ACID properties of transaction, thus it cannot support the strong transaction properties as the relational database does. In addition, the majority of No SQL databases are based on the KV strategy, thus the range query is not quite efficient. Moreover, each No SQL database uses the different syntax, which does not support the SQL language.Here, I will introduce the research about some key technologies of memory database, mainly describing the index technology, transaction and concurrency control technology. This paper also discussed how to apply these technologies to the relational database. In this paper, the main parts of this work are as follow:(1)Analyzing database index technology. Inspired by string suffix array, we came up with B+ tree index that is based on suffix array. In this case, we combined the existing B+ tree structure with the suffix array. This strategy improves the efficiency of fuzzy matching in which the string has before and after percent sign, instead of full table traversal used in the traditional relational database.(2)In order to optimize processing the tree-like data in the database, we designed an index strategy named HB tree. This strategy has high performance in traverse up and down the tree, and can access to every node in the tree-like data quickly. Besides, the HB tree can combine with other indexes in the database and optimize the query efficiency. This strategy improves the performance of database in the file system.(3) In the aspect of transaction and concurrency control, we used the existing rollback area, and came up with a concurrency control strategy based on the timestamp management. In this way, lock is not required in query. Thus, it improves the concurrency of reading and simplified the locking strategy. Furthermore, we designed an asynchronous persistence technology that keeps the data persistent block by block. This strategy improves the performance of persistence.(4)Based on the research on database technology, we designed and developed a small memory database. This database is programmed by C++,and has been optimized but need to be further tested.