Construction and validation of a reconfigurable computer cluster.
Date
Authors
Access rights
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Cluster computing networks multiple computers (nodes) to exploit their parallel processing power. By pooling resources of multiple computers, computations can be decomposed and allocated across nodes with partial solutions collected and combined to form a complete solution (scatter and gather). By augmenting cluster nodes with reconfigurable hardware, systems can be configured with assistive devices that improve their computational performance; thus reducing computation time and increasing computational flexibility. This thesis evaluates a cluster of 16 Virtex II Pro Development boards that were integrated as an experimental cluster. The well-known 3DES algorithm was used to measure the runtime of multiple partitioned datasets (1 to 16 partitions) to quantify the execution speedup over a varying number of nodes. The results showed that performance can be improved with hardware acceleration, although there is complex interplay between node communication and file I/O timing that impacts the magnitude of the speedup.