Slowly Changing Dimensions Research Articles

Design and implementation of ETL processes using BPMN and relational algebra

Extraction, transformation, and loading (ETL) processes are used to extract data from internal and external sources of an organization, transform these data, and load them into a data warehouse. The Business Process Modeling and Notation (BPMN) has been proposed for expressing ETL processes at a conceptual level. A different approach is studied in this paper, where relational algebra (RA), extended with update operations, is used for specifying ETL processes. In this approach, data tasks in an ETL workflow can be automatically translated into SQL queries to be executed over a DBMS. To illustrate this study, the paper addresses the problem of updating Slowly Changing Dimensions (SCDs) with dependencies, that is, the case when updating a SCD table impacts on associated SCD tables. Tackling this problem requires extending the classic RA with update operations. The paper also shows the implementation of a portion of the TPC-DI benchmark that results from both approaches. Thus, the paper presents three implementations: (a) An SQL implementation based on the extended RA-based specification of an ETL process expressed in BPMN4ETL; and (b) Two implementations of workflows that follow from BPMN4ETL, one that uses the Pentaho DI tool, and another one that uses Talend Open Studio for DI. Experiments over these implementations of the TPC-DI benchmark for different scale factors were carried out, and are described and discussed in the paper, showing that the extended RA approach results in more efficient processes than the ones produced by implementing the BPMN4ETL specification over the mentioned ETL tools. The reasons for this result are also discussed.

Content enrichment with expressive document modelling to leverage the understanding of unstructured data

Most information in an enterprise is in the form of unstructured data which is usually managed using a document database. One of the key challenges is to define a generalized data model for this unstructured data and any information extracted from it using content enrichment algorithms. It is more challenging to incorporate provenance and temporal capabilities to such data models. Semantic databases use ontologies such as PROV-O to represent their provenance information expressively, and relational databases use for example Slowly Changing Dimensions (SCDs) concepts to represent temporal information. In this paper, we present a document model which has features inspired from Dublin core, PROV-O and temporal methodologies to generalize information extracted from unstructured data using content enrichment algorithms. Provenance information enables comparison of enrichment models, allows reproducibility and facilitates complex filtering on the enriched data. Temporal metadata helps in versioning the document and enables point-intime and history queries conveniently.

Handling slowly changing dimensions in data warehouses

Analysis of historical data in data warehouses contributes significantly toward future decision-making. A number of design factors including, slowly changing dimensions (SCDs), affect the quality of such analysis. In SCDs, attribute values may change over time and must be tracked. They should maintain consistency and correctness of data, and show good query performance. We identify that SCDs can have three types of validity periods: disjoint, overlapping, and same validity periods. We then show that the third type cannot be handled through the temporal star schema for temporal data warehouses (TDWs). We further show that a hybrid/Type6 scheme and temporal star schema may be used to handle this shortcoming. We demonstrate that the use of a surrogate key in the hybrid scheme efficiently identifies data, avoids most time comparisons, and improves query performance. Finally, we compare the TDWs and a surrogate key-based temporal data warehouse (SKTDW) using query formulation, query performance, and data warehouse size as parameters. The results of our experiments for 23 queries of five different types show that SKTDW outperforms TDW for all type of queries, with average and maximum performance improvements of 165% and 1071%, respectively. The results of our experiments are statistically significant.

MapReduce-based dimensional ETL made easy

This paper demonstrates ETLMR , a novel dimensional Extract--Transform--Load (ETL) programming framework that uses Map-Reduce to achieve scalability. ETLMR has built-in native support of data warehouse (DW) specific constructs such as star schemas, snowflake schemas, and slowly changing dimensions (SCDs). This makes it possible to build MapReduce-based dimensional ETL flows very easily. The ETL process can be configured with only few lines of code. We will demonstrate the concrete steps in using ETLMR to load data into a (partly snowflaked) DW schema. This includes configuration of data sources and targets, dimension processing schemes, fact processing, and deployment. In addition, we also present the scalability on large data sets.