Towards a dynamic parallel database machine: data balancing techniques and pipeline

Abstract

(eng) The fast development over the last years of high performance multicomputers makes them attractive candidates as the base technology for scalable and performance oriented database applications. In this paper, we address the problem of how to process utility commands while the system remains operational and the data remain available for concurrent access. In particular, we focus on the on-line reorganization of a dictionary, a database reduced to its simplest instance, showing its implementation on a multicomputer. As is the case with implementations of dynamic structures on distributed memory architectures, a crucial load balancing problem has to be solved. We propose an elegant solution and prove that it solves this problem. Experimental results are shown and analyzed.