The Design And Implementation Of Distributed Database Middleware DBScale

It’s more and more difficult for traditional stand-alone relational database to process growing big data. This paper will introduce a distributed database middleware DBScale based on MySQL which is most popular open source relational database. The distributed database middleware can be used just like MySQL and the original application program do not need to modify any code which only need to migrate data from MySQL to DBScale.This paper will study the distributed database middleware, including three modules. The first is the schema spilted and table splited. The second is cross node join. The last is the distributed stored procedure. It must to implement the table splited efficiently for each distributed database middleware. For the purpose of easy scalability, this paper uses consistency hash algorithm and virtual partition to implement the table splited. The consistency hash algorithm can guarantee migrate the least data when add a server in distributed database middleware. Cross node join is a ver hard problem. This paper is divided into three cases to discuss including two norm table joining, two partition table joining and one norm table joinging one partition table. No matter what kind of situation the core idea of this article is to try to reduce the amount of data migration. Finally the paper will introduce the distributed stored procedure based on the standard MySQL stored procedures, including statements expression, complex expression, circulation expression, and conditional branch judgment and cursor expression.Based on the table spilted test, we found that as the amount of back end MySQL is increasing the performance of DBScale will be a linear growth. Based on the corss node join test, we found the performance of cross node join will be 50% of the MySQL when using Federated storage engine. Based on the comparison test with cobar and ha-proxy, we found that the performance of DBScale is better than cobar each time. While the performance of DBScale is better than ha-proxy under the high concurrency...