In recent years, with the rapid popularization of the embedded devices in the financial, communications, electronics, aerospace, industrial automation control, medical and other fields, the amount of data also continue to grow in embedded application environment, therefore it has become particularly important to provide data management for embedded systems. At present, the embedded real-time database has become a hot research field. Traditional database doesn't consider the time limitation of data and transaction, whose object is average performance and system throughout, while the performance standard to measure the embedded real-time database systems is real-time transaction success rate and system resource utilization. Therefore, we must research a new transaction processing technology for embedded real-time database.The main contents done in this paper are as follows. First, we analyze the factors of real-time transaction execution time prediction. Based on the analysis, a feasible transaction execution time estimation strategy is given, and we verify the validity of the strategy through testing. Second, we analyze the advantages and disadvantages of existing transaction priority assignment strategy. Based on the analysis, an optimal transaction scheduling strategy is proposed and the implementation in embedded real-time database system CS_ERTDB developed by ourselves is also given. Third, we design and implement the concurrency control module of CS_ERTDB, including the design of the lock table structure which integrates the functions of transaction management, lock management and recovery management. Finally, we investigate the problem of traditional transaction recovery strategy and propose a automatic recovery strategy which can use the other concurrent transactions to recover when one transaction failure. The experiment demonstrates that the strategy has good performance in reducing downtime and improving the success rate of transaction. |