Features of efficient processing of SQL-queries to conservative type databases

In article, we are considering the problems of processing SQL queries with a high specific weight of join operations to conservative type databases (with occasional updating of data in specially allocated time) with increased data volumes on the GPU-cluster platform. Modified architecture of the previously developed parallel DBMS Clusterix-N (N - from New). We are proposed the methods for organizing dynamic segmentation of intermediate / temp relationships on a dedicated node with GPU accelerators and full load of processor cores of nodes on JOIN and IO levels. The performance of this DBMS is compared with the original DBMS PerformSys on a limited TPC-H test (without write operations) with V$_{\mathrm{DB}}$=60 GB and V$_{\mathrm{DB}}$=120 GB.