Relatively inexpensive hybrid technology of large volumes conservative DBMS

In article are discussed the issues of organization of conservative DBMS on comparatively inexpensive cluster platforms using MySQL and GPU-accelerators at the executive level. The relevance of the adopted orientation for work with large-scale databases is determined by current trends in the intellectual processing of large information arrays. Increasing the size of databases requires them to be hashed over cluster nodes. This necessitates the use of a regular query plan. The use of homogeneous cluster technologies (DBMS Clusterix with the optimal configuration "combined symmetry") requires an additional dynamic segmentation of intermediate and temporal relationships. Hybrid technologies (Clusterix-N and Clusterix- G projects) are proposed for managing large-scale databases, which allows excluding dynamic segmentation and significantly improving efficiency by the criterion of productivity / cost. In them, the data is located in nodes RAM. A distinctive feature of the Clusterix-G DBMS with GPU accelerators is the work with compressed databases, which allows increase their data volume on nodes with limited amount of memory. The proposed technologies are more efficient in comparison with lusterix, and the Clusterix-G performance should exceed Clusterix-N for medium speed interconnect.