Digital Library

cab1

 
Title:      GENERATING ARTIFICIAL LOAD PATTERNS ON MULTI-PROCESSOR PLATFORMS
Author(s):      Turkey Alsalkini, Greg Michaelson
ISBN:      978-989-8533-25-8
Editors:      Hans Weghorn
Year:      2014
Edition:      Single
Keywords:      Load, Multi-processor, Multi-core, Evaluation, Balancing, Mobile Computing
Type:      Full Paper
First Page:      77
Last Page:      84
Language:      English
Cover:      cover          
Full Contents:      click to dowload Download
Paper Abstract:      It is desirable to evaluate system performance under repeatable conditions. Typically, parallel systems are evaluated on dedicated platforms with little or no external impact on load. For dynamic systems, however, such as those that adapt to changing conditions, it is necessary to generate both predictable and realistic patterns of load. Furthermore, on shared non-dedicated systems, there is a need to compensate for the unpredictable loads introduced by other users. We have developed a novel load function which may be instantiated to generate predictable patterns of load across multiple processors. Our function can both generate idealised load patterns, and record and playback real load patterns. Furthermore, it can dynamically maintain a required load pattern in the presence of external real-time load changes, which makes it particularly suitable for experimentation on shared systems. We discuss the design of the load function and show that it can generate dynamic, adaptive and precise load, with minimal impact on system load. We then illustrate its use in the experimental evaluation of static/dynamic load balancing, load stealing and mobile skeletons.
   

Social Media Links

Search

Login