Data Intensive Systems (DIS) Benchmark Performance Summary
Abstract
Peak processor performance increases at a rate of 60% per year, but memory access speeds increase at a rate of only 7% per year. Computing-system designers compensate for the resulting divergence by incorporating caches or latency-hiding measures into their designs. However, elements such as larger caches, prefetching, and multithreading do not address the needs of data-intensive DoD applications, which consequently operate at rates far below the peak processor- capacity. As the mismatch between processor and memory grows the number of applications unable to operate at peak rates increases. The DARPA Data Intensive Systems Program was created to address this problem. A variety of novel architectures or enhancements were developed under this program to increase the effective performance-as opposed to the rated peak performance of systems running data-starved applications. Under this project, a DIS Benchmark Suite was developed to measure the performance of the prototypical systems. Additionally, the DIS Stressmark Suite was developed to assist performance measurement during the development process. Participating teams were expected to utilize these tools and supply their measurements. In this report the benchmarking tools are introduced, the reported results are summarized, and an objective analyses of the results is provided.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 2003
- Accession Number
- ADA418752
Entities
People
- Joseph Musmanno