Constructive Modeling of Conservative DBMS
Vadim A. Raikhlin, Roman K. Klassen · 2019 International Russian Automation Conference (RusAutoCon) · 2019
Commercial OLAP-systems are quite expensive and therefore inaccessible to organizations with limited financial capabilities. Analytical processing of significant amounts of data in these organizations can be carried out using open source software systems on low price cluster platform. Previously made Clusterix-like conservative-type DBMSs (with occasional data updates) using a regular query processing plan were not sufficiently effective. Therefore, research on such systems has been developed with a focus on the full load of processor cores and the use of dynamic segmentation of intermediate and temporary relations on the GPU (Clusterix-N, N - from New) up to the development of a system comparable in efficiency to the promising open Spark system. The development methodology was based on the constructive system modeling methodology. According to this methodology, the simulation of the synthesis process of the considered DBMS was carried out iteratively. Each iteration involved a complete software development of the corresponding state of the model and analysis of the results of testing on the GPU cluster platform. The required simulation quality was obtained at the fourth iteration.