Economical and superior database selection for skewed data warehouses

Since the 1970s, RDBMS databases have been the dominant choice, not only for OLTP systems but also for analytical applications. This dominance is driven by both technical factors and the historical evolution of database systems based on emerging needs. Before the 1980s, analytical requirements were primarily handled directly through OLTP systems. However, after the 1980s, analytical databases emerged as standalone systems based on Inmon and Kimball models, offering an alternative to the traditional 3NF approach and operating on separate platforms. Although columnar database applications have existed since as early as 1969 [1,2], they did not gain significant popularity until the last 15 years. Their growing adoption is largely due to the exponential increase in data volume, which continues to expand rapidly. Thanks to their unique architecture, columnar databases have demonstrated superior performance in data warehouse (DWH) solutions. This thesis first investigates whether columnar databases outperform traditional RDBMS systems. If they do, we aim to determine the extent of their superiority, the conditions under which they perform better, and the key factors contributing to their improved performance. Secondly, we address the cost aspect of this comparison. Since our research is concerned with real-world business applications, we will evaluate the financial implications of using columnar databases compared to RDBMS solutions. There should be a correlation between their cost and performance gains; ideally, columnar databases would not only offer better performance but also come at a lower cost than RDBMS, which would be the preferable outcome if achievable. Another critical aspect of our study is examining a real-world DWH implementation to ensure a practical and comprehensive comparison. In business environments, various exceptions and complexities make it difficult to validate theoretical results. To address this, we analyze a GSM company's DWH, which utilizes dimensional modeling (specifically the constellation/galaxy subtype) rather than relying solely on the Star Schema Benchmark (SSBM) [3]. Keywords: (Columnar, RDBMS, database, DWH, MPP) Numeric Code of the Field: 92432