High-Speed, Low-Cost Workstation for Computation-Intensive Statistics. Phase 1
Abstract
High-performance and low-cost workstations are essential for economical solution of computationally intensive statistical problems. This program evaluated the use of a digital signal processor (DSP) as a number- cruncher in these applications to achieve a significant improvement in speed at an acceptable cost. To demonstrate the feasibility of this approach, a number of representative statistical procedures were selected. These statistical procedures were examined in detail, their common functions were identified, and their hierarchical structure determined. A set of DSP-based mathematical and statistical subroutines were coded and optimized for speed. These subroutines were then incorporated in the selected statistical application programs. Benchmark evaluation was performed using a commercial DSP peripheral board, interfaced to a personal computer. A total of 62 low level software routines and 15 diverse statistical applications were benchmarked. The feasibility of the proposed approach was conclusively demonstrated, with 100 to 200 times computation speed improvements over a standard 386-type personal computer operating at 20 MHz clock frequency. There are many potential applications of a cost-effective workstation in industrial, government, and academic settings. The proposed statistical workstation will be designed to meet the need of these users. Digital signal processing, Statistics, Mathematics, Computational, Simulation, Personal computer, Workstation, Hardware accelerator, DSP, Software.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 20, 1990
- Accession Number
- ADA275153
Entities
People
- Hongru Shen
- Janis Pukite
- Michael J. Kernal
- Paul R. Pukite