H-DB: Yet Another Big Data Hybrid System of Hadoop and DBMS

摘要

With the explosion of the amount of data, analytics applications require much higher performance and scalability. However, traditional DBMS encounters the tough obstacle of scalability, and could not handle big data easily. In the meantime, due to the complex relational data model, the large amount of historical data and the independent demand of subsystems, it is not suitable to use either shared-nothing MPP architecture (e.g. Hadoop) or existing hybrid architecture (e.g. HadoopDB) to replace completely. In this paper, considering the feasibility and versatility of building a hybrid system, we propose a novel prototype H-DB which takes DBMSs as the underlying storage and execution units, and Hadoop as an index layer and a cache. H-DB not only retains the analytical DBMS, but also could handle the demands of rapidly exploding data applications. The experiments show that H-DB meets the demand, outperforms original system and would be appropriate for analogous big data applications.