Physical modeling of data warehouses using UML

Abstract

During the few last years, several approaches have been proposed to model different aspects of a Data Warehouse (DW), such as the conceptual model of the DW, the design of the ETL (Extraction, Transformation, Loading) processes, the derivation of the DW models from the enterprise data models, etc. At the end, a DW has to be deployed to a database environment and that takes many decisions of a physical nature. However, few efforts have been dedicated to the modeling of the physical design (i.e. the physical structures that will host data together with their corresponding implementations) of a DW from the early stages of a DW project. From our previously presented DW engineering process, in this paper we present our proposal for the modeling of the physical design of DWs by using the component diagrams and deployment diagrams of the Unified Modeling Language (UML). Our approach allows the designer to anticipate important physical design decisions that may reduce the overall development time of a DW such as replicating dimension tables, vertical and horizontal partitioning of a fact table, the use of particular servers for certain ETL processes and so on. Moreover, our approach allows the designer to cover all main design phases of DWs, from the conceptual modeling phase until the final implementation, as we show with an example in this paper.

A UML profile for multidimensional modeling in data warehouses

“…of a commercial OLAP tool to query the implemented multidimensional model. For more details on how to implement other data warehouse parts, we refer the reader to [22,21,23,36].…”

“…-A StarPackage can only contain FactPackages or DimensionPackages: 22 self.contents -> forAll(me j me.ocllsTypeOf(FactPackage) or me.ocllsTypeOf(DimensionPackage)) -A StarPackage can only contain one FactPackage: 23 self.ownedElement -> select(me j me.ocllsTypeOf(FactPackage)) -> size <= 1 -A StarPackage cannot import a FactPackage from another StarPackage (only DimensionPackage):…”

Applying MDA to the development of data warehouses

“…In our approach we use UML [17] as the modeling language for constructing PIMs and PSMs, since we can use its extension mechanisms to adapt it to specific domains. In previous works several UML profiles have been developed to adapt UML to certain aspects of DW design: MD modeling [9,10,22], modeling of the ETL processes [23], modeling data mappings between data sources and targets [12], or physical modeling of the DW [13]. Furthermore, some metamodels from CWM [18] can be used to model PSMs.…”

“…To overcome these problems, in a previous work [11] a DW development method based on the UML (Unified Modeling Language) [17] and the UP (Unified Process) [5] has been defined. This method addresses the design of the whole DW from operational data sources to the final implementation by using several UML profiles which have been developed to adapt UML to certain aspects of the DW design, such as MD modeling [9,10,22], modeling of ETL processes [23], modeling data mappings between data sources and targets [12], and physical modeling of the DW [13]. This method has a clear benefit: using the same standard notation (i.e.…”

A UML Profile for Representing Business Object States in a Data Warehouse

“…Trujillo and Luján-Mora introduced a Data Warehouse design framework in [7], which is supported among others by the UML Profile for Modeling ETL processes [8], a Profile for Physical Modeling of Data Warehouses [9], and a Profile for Multidimensional Modeling [10].…”