Data Exchange For Real-Time Main Memory Database

In real-time databases, it is desirable to execute transactions within their deadlines using fresh (temporally consistent) data which can reflect the continuously changing external environments, e.g., current temperature or stock prices. Traditional databases can not support time constraints and data temporal consistency for disk I/Os. While in MMDBs(Main Memory Database), most of disk I/Os has been removed during the normal processingTraditional MMDBs assume that the entire databases are stored in volatile main memory, while their backup copies are kept in an Archive Database (AD) residing on secondary storage. This approach can not work when the volatile main memory is not large enough, especially in embedded systems with limited main memory. The appealing alternative is just to guarantee that no disk I/Os block a transaction when running it. In this paper, initialization and data exchange schemes are proposed to realize the new definition of MMDB with limited main memory, based on which an ARTs-EDB is realized. Its performance indicates that the initialization and data exchange scheme adapts well to the limited main memory.The scheme takes transaction priority and data characteristics such as temporality and access frequency into account for the data exchange. Most of the scheme is based on the result of static pre-analysis of transaction for ARTs-EDB, which distills the time information, data characteristics, and access processes of the transaction. The "super access process" makes a transaction started before its whole dataset placed in ARTs-EDB, reducing the probability of missing deadline. This new approach greatly improves the practicability of MMDBs to embedded real-time database.