Mapddb: A Distributed Relational Database System with row-based keywords

Mapddb: A Distributed Relational Database System with row-based keywords. It is mainly used to solve horizontal expansion of the database system. In a real Internet business application system, the database processing capability of the original system cannot keep up with the growth of the business scale, generally, you can only constantly upgrade the hardware system, such as upgrading a PC server to a minicomputer, upgrading a minicomputer to a medium or large machine, or using the multi-node mode that comes with the database system, if Oracle is used, you can use RAC + San. However, in a RAC system, the number of nodes generally does not exceed 4. After the number is exceeded, the number of nodes is increased, and the performance is not increased much. Therefore, this method is still limited. Therefore, based on the characteristics of most Internet business applications, we can scale out a distributed database system (mapddb) by adding nodes to cope with business growth. The principle of this system is that the data in the database can be split by a certain row keyword. For example, the business data in the instant messaging system can be split by user account, because the contact and other data belong to a specific account, you can perform database sharding according to the user account. In order to solve the problem where user data is located in the sub-database, add a set of management line keywords and the corresponding relational system of the database. In this way, when the applicationProgramTo access the data of an account, first find the corresponding sub-database based on the corresponding relationship, and then directly access the database to access the data.

At present, this system has been successfully applied to the Alibaba trademanager System of Alibaba software. It has abandoned the performance bottleneck problem caused by the previous use of minicomputers and greatly improved the response time and scalability of the Alibaba trademanager system, this greatly reduces costs and replaces the minicomputer + San structure with PC servers. At present, there are several sets of mapdd in the application. There is a set of six nodes and more than 30 billion records are stored.