Designing parallel relational data warehouses: A global, comprehensive approach

Abstract

The process of designing a parallel data warehouse has two main steps: (1) fragmentation and (2) allocation of so-generated fragments at various nodes. Usually, we split the data warehouse horizontally, allocate fragments over nodes, and finally balance the load over the nodes of the parallel machine. The main drawback of such design approach is that the high communication cost. Therefore, Data Replication (DR) has become a requirement for availability on the one hand but also for minimizing the communication cost on the other hand. In this paper, we present a redundant allocation algorithm for designing shared-nothing parallel relational data warehouses, which is based on the well-known fuzzy k-means clustering algorithm.