Performance Evaluation of Multiple Processor Systems.
Abstract
Technological trends in semiconductor and micro-processor development are clearly leading towards the production of a computer on a chip. In the near future such a computer module will include the equivalent of today's mini-computer with some memory. Connecting many computer-modules together will probably be a cost-effective way to build high performance computer systems. This thesis investigates and contrasts the performance of two multiple computer-module structures: a multiprocessor with shared memory and a local computer network in which all communication is via messages. Results are derived both from running benchmark applications and from performance models. Very little knowledge and experience has been gained on the performance of actual multiple processor applications. To investigate the problems and potential of these computer-module structures, a multiple processor system -- Cm** -- has been built. Firmware changes enable the emulation of both an efficient multiprocessor and a local computer network. Experiments were conducted on both types of structures to obtain performance information. Out practical methodology includes measurement of performance parameters using a set of benchmark application programs executing on Cm**, and performance models that were derived and validated - using the measurement results - and then applied for the performance investigation.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1978
- Accession Number
- ADA060495
Entities
People
- Levy Raskin
Organizations
- Carnegie Mellon University