skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: A Quantitative Measure of Memory Reference Regularity

Abstract

The memory performance of applications on existing architectures depends significantly on hardware features like prefetching and caching that exploit the locality of the memory accesses. The principle of locality has guided the design of many key micro-architectural features, including cache hierarchies, TLBs, and branch predictors. Quantitative measures of spatial and temporal locality have been useful for predicting the performance of memory hierarchy components. Unfortunately, the concept of locality is constrained to capturing memory access patterns characterized by proximity, while sophisticated memory systems are capable of exploiting other predictable access patterns. Here, we define the concepts of spatial and temporal regularity, and introduce a measure of spatial access regularity to quantify some of this predictability in access patterns. We present an efficient online algorithm to dynamically determine the spatial access regularity in an application's memory references, and demonstrate its use on a set of regular and irregular codes. We find that the use of our algorithm, with its associated overhead of trace generation, slows typical applications by a factor of 50-200, which is at least an order of magnitude better than traditional full trace generation approaches. Our approach can be applied to the characterization of program access patterns and in themore » implementation of sophisticated, software-assisted prefetching mechanisms, and its inherently parallel nature makes it well suited for use with multi-threaded programs.« less

Authors:
; ; ; ;
Publication Date:
Research Org.:
Lawrence Livermore National Lab., CA (US)
Sponsoring Org.:
US Department of Energy (US)
OSTI Identifier:
15006306
Report Number(s):
UCRL-JC-145580
TRN: US200407%%202
DOE Contract Number:  
W-7405-ENG-48
Resource Type:
Conference
Resource Relation:
Conference: International Parallel and Distributed Processing Symposium 2002, Fort Lauderdale, FL (US), 04/15/2002--04/19/2002; Other Information: PBD: 1 Oct 2001
Country of Publication:
United States
Language:
English
Subject:
99 GENERAL AND MISCELLANEOUS//MATHEMATICS, COMPUTING, AND INFORMATION SCIENCE; ALGORITHMS; DESIGN; IMPLEMENTATION; PERFORMANCE; PROCESSING

Citation Formats

Mohan, T, de Supinski, B R, McKee, S A, Mueller, F, and Yoo, A. A Quantitative Measure of Memory Reference Regularity. United States: N. p., 2001. Web.
Mohan, T, de Supinski, B R, McKee, S A, Mueller, F, & Yoo, A. A Quantitative Measure of Memory Reference Regularity. United States.
Mohan, T, de Supinski, B R, McKee, S A, Mueller, F, and Yoo, A. Mon . "A Quantitative Measure of Memory Reference Regularity". United States. https://www.osti.gov/servlets/purl/15006306.
@article{osti_15006306,
title = {A Quantitative Measure of Memory Reference Regularity},
author = {Mohan, T and de Supinski, B R and McKee, S A and Mueller, F and Yoo, A},
abstractNote = {The memory performance of applications on existing architectures depends significantly on hardware features like prefetching and caching that exploit the locality of the memory accesses. The principle of locality has guided the design of many key micro-architectural features, including cache hierarchies, TLBs, and branch predictors. Quantitative measures of spatial and temporal locality have been useful for predicting the performance of memory hierarchy components. Unfortunately, the concept of locality is constrained to capturing memory access patterns characterized by proximity, while sophisticated memory systems are capable of exploiting other predictable access patterns. Here, we define the concepts of spatial and temporal regularity, and introduce a measure of spatial access regularity to quantify some of this predictability in access patterns. We present an efficient online algorithm to dynamically determine the spatial access regularity in an application's memory references, and demonstrate its use on a set of regular and irregular codes. We find that the use of our algorithm, with its associated overhead of trace generation, slows typical applications by a factor of 50-200, which is at least an order of magnitude better than traditional full trace generation approaches. Our approach can be applied to the characterization of program access patterns and in the implementation of sophisticated, software-assisted prefetching mechanisms, and its inherently parallel nature makes it well suited for use with multi-threaded programs.},
doi = {},
journal = {},
number = ,
volume = ,
place = {United States},
year = {2001},
month = {10}
}

Conference:
Other availability
Please see Document Availability for additional information on obtaining the full-text document. Library patrons may search WorldCat to identify libraries that hold this conference proceeding.

Save / Share: