With the rapid development of computer technology, the application field ofmodern database is expanding wider and wider. As main database are all on diskettesin relational database system based on disk, transaction processing is associated withdisk I/O operation, which is difficult to meet high performance data accessingcapability for application systems. A main memory database(MMDB) which supportsthe real-time transaction as rock-bottom provides a well solution of real-time datamanagement for some application areas. It is very important to support real-timedatabase system.This paper analyze the rock-bottom of real-time database system, which arearchitecture of main memory database, data accessing, data organization andtransaction management. The conception of main memory database is broughtforward, and the difference between main memory database and conventional diskdatabase is analyzed. A four-leveled storage architecture is proposed. Thecharacteristics of real-time data and transactions is discussed. On this basis, atransaction scheduling algorithm which consider time delay is broughtout. Experiment results show the shortest transaction is chosen to be scheduled, whichcan provide significant gains over traditional scheduling algorithms in real-timedatabase.On the other hand, as main memory is volatile memory and database can bedestroyed easily, backup and recovery techniques play an important role in MMDBperformance. This paper gives a model for data recovery on the combination of FuzzyCheckpointing technology and Logging After Write log protocol. It can cut down theamount of recovery work, shorten recovery time and improve the efficiency ofrecovery. Experiment results show that the model is feasible. |