A Statistical-Empirical Hybrid Approach to Hierarchical Memory Analysis
Authors: X.-H. Sun, K. Cameron
Date: September, 2000
Venue: The Euro-Par 2000, Lecture Notes in Computer Science 1900, Springer
Type: Conference
Abstract
A hybrid approach that utilizes both statistical techniques and empirical methods seeks to provide more information about the per- formance of an application. In this paper, we present a general approach to creating hybrid models of this type. We show that for the scientific applications of interest, the scaled performance is somewhat predictable due to the regular characteristics of the measured codes. Furthermore, the resulting method encourages streamlined performance evaluation by determining which analysis steps may provide further insight to code performance.