Intelligent text recognition system on a heterogeneous multi-core processor cluster: A performance profile and architecture exploration

High Performance Computing (HPC) clusters using multi-core processors requires extremely well designed software in order to fully take advantage of the power presented by the system. In order to evaluate the software's performance on the cluster, it is helpful to collect a performance profile of engineering data gathered from a wide range of test inputs. The performance profile can be used to help analyze the strengths and weaknesses of the software implementation.;The goal of this thesis work is to build a performance profile of the Intelligent Text Recognition (ITR) software running on a cluster of IBM Cell Broadband Engine Processors at the Advanced Microelectronics and Power-Aware Systems (AMPS) Lab at Binghamton University. The ITR software is based upon a Hybrid Cognitive Model which utilizes two Artificial Neural Networks: the Brain-State-in-a-Box and Confabulation algorithms. The performance profile is used to carry out an architecture exploration which investigates the optimal software hierarchy and process mapping of the artificial neural networks utilized in the ITR system.