Design And Implementation Of Distributed Database Middleware Based On Deterministic Execution Strategy

With the rapid development of storage and data amount in computer system,distributed database system has became the main form of data storage in most of enterprises. In order to improve the ability of data processing, the distributed database needs to add the new node to improve the performance of the whole database. In order to ensure the ACID property of transaction, the distributed database need to implement the two phase of the protocol. Considering the two phase commit protocol transforming the data for many times during the network, the delaying time of the network will surpass the transaction execution time in local database. When the number of nodes in a distributed database exceeds a certain number, the performance of the entire distributed database will be gradually reduced until maintained in a stable level. But the handling capacity will not grow linearly as the number of database node. Maintaining the ACID attributes of the distributed transaction have become the main reason to limit the scalability of the distributed database. Abadi Daniel, Professor of Yale University, proposed the concept of deterministic transaction execution strategy in 2010 that if each distributed database node receives requests sorted in certain global order and eliminates the randomness of the system thread scheduling in each node and the possibility of a user to modify or break a transaction in the course of a transaction. The distributed database system can ensure the consistency of the database copy without using the two phase commit protocol.This paper designed and implemented a database middleware which can make relational database into a deterministic strategy of distributed database that will own excellent feasibility with the ACID transaction property. In this paper, the following work is done:(1) Global ordering of transaction requests. A data communication module is designed and implemented in this paper. In the data communication module, using Appia to encapsulate the information exchange of different database middleware into a Ddchannel. Using the information exchange between Ddchannel to determine the final transaction request execution sequence to make a transaction request in the entire distributed database to achieve a global order.(2) Eliminating the randomness of the system thread scheduling in each distributed database node and eliminating the possibility of a user modifying or breaking a transaction in the course of a transaction. In this paper, a transaction management module is designed and implemented. The transaction is applied by the conservative ordered lock to eliminate the uncertainty of operating system thread scheduling. The transaction management module invokes the stored procedure to execute a transaction request in order to eliminate the possibility of a user to modify or break a transaction in the course of a transaction.(3) Verifying the performance of the final implementation of the database. In this paper, we test the performance of read and write performance test and transaction throughput test for the final realization of the database. The results show that the final implementation of the database achieved in the paper has a good linear scalability and extensibility.