The DiNG - A Parallel Multiattribute File System for Deductive Database Machines

We investigate the main requirements for mass storage subsystems with regard to the needs of deductive database machines. In particular, we focus in this paper . on fully integrated index data structures, namely a multiattribute search structure based on a grid file derivate and . on certain hardware requirements, i.e., redundant arrays of inexpensive disks in order to avoid a disk I/O bottleneck during rule processing. The former requirement is motivated by an example, in particular by a closer look at the execution of Datalog-* rule sets. As a solution, we propose the DiNG file system as speciaiised mass storage subsystem for high performance deductive database machines. The DiNG file system is a highly parallel me system which supports symmetric multiattribute queries against factand rulebases. At the moment, the DiNG file system is operational on an experimental shared nothing MIMD machine, namely an eightprocessor/eight-disk intel iPSC/2.