Research On Key Evaluating Techniques Of HBase Database

With the development and wide application of Internet technology, the volume of information systems is growing rapidly. The traditional relational database has been unable to meet the needs of the emerging business. In response to the massive data storage and management needs, the major manufacturers and scientific research institutions have launched various No SQL storage solutions. HBase database is one of the most famous solutions. However, with the popularization and promotion of HBase database, how to evaluate HBase database has become a hot spot of academic research. In order to achieve performance analysis and tune of HBase database, this paper simulates real production environment workload and failure in the evaluation. It is of great significance to the development of HBase database. Two directions of evaluating HBase were presented. One is performance test. Another is fault-tolerant test. An evaluation platform for HBase database was desiged and implemented.In the performance test, a distributed performance test tool was constructed according to the characteristics of the HBase database. Also a metric system of performance evaluation was put forward. The tool uses the classic server-client distributed structure, to avoid the problem that the load generator may become the bottleneck of the system, which can be deployed on cheap PCs. In the implementation process several technical difficulties have been broke throgh, eg. the data publish and subscribe, distributed mutex, distributed barricades.In fault-tolerant tests, the fault injection mode was used to evaluate the fault-tolerant ability of HBase database. Also the unique architecture of HBase was considered. In order to simulate the real fault of HBase, this paper has considerd to inject fault from four different layers, process layer, component layer, network layer and function layer. A fault injection tool for HBase was implemented.Finally, a small HBase database system which consists of four servers has been set up. The distributed performance testing tools and fault injection tool were integrated for HBase database performance evaluation. Also the corresponding experimental results were analyzed. Several performance improvement recommendations were given. And the correctness of the performance evaluation platform was verfied.