Návrh metodického postupu tvorby multidimenziálního datového modelu v systému Business Inteligence

Abstract

Business Intelligence (BI) system is information system’s key subsystem that provides access to accurate, timely and relevant information for the support of decision-making in the management. BI system’s database quality and usefulness is substantially affected mainly by changes in user requirements. It is therefore imperative to apply a methodical process of the design that ensures highest level of flexibility and resistance to this phenomenon. The first goal is to propose a methodical process of BI system’s data model design that is based on principles of the agile oriented methodology Scrum, as a framework for effective process of the multidimensional data model analysis and design. In the thesis there is an example case study of the application of the proposed methodical process presented. The second goal is to assess applicability of the anchor modeling technique in the context of the multidimensional data model implementation. It is a relational modeling technique which is also compatible with the methodical process as potential method of delivering the output of the process. The technique offers also tools for achieving flexibility in the BI system’s data model development process and that are compatible with the proposed methodical process’ philosophy. The applicability of the anchor modeling technique is analyzed in context of a standard approach which is based on the construction of a non-normalized star schema. Two typical multidimensional data models are used with unique sets of 250 star queries for each of the models. Statistical significance of query execution results’ differences is tested using paired samples comparison methods (at the 5 % level of significance). Query processing characteristics, in terms of selected physical operators’ performance, are used to explore relationships with the resulting query output size, using the multivariate analysis of variance method. Results of the analysis are then used during the detailed analysis of query performance results and query content. Conclusions of the thesis present indication of observed query contents and parameters that constrain the query execution performance when using the anchor modeling based approach to construct the multidimensional data model. The conclusions also indicate potential constraints for the use of the approach as an output delivery method for the methodical process.