A hybrid strategy based on data distribution and migration for optimizing memory locality

I. Kadayif, Mahmut Kandemir, A. Choudhary

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

The performance of a NUMA architecture depends on the efficient use of local memory. Therefore, software-level techniques that improve memory locality (in addition to parallelism) are extremely important to extract the best performance from these architectures. The proposed solutions so far include OS-based automatic data migrations and compiler-based static/dynamic data distributions. This paper proposes and evaluates a hybrid strategy for optimizing memory locality in NUMA architectures. In this strategy, we employ both compiler-directed data distribution and OS-directed dynamic page migration. More specifically, a given program code is first divided into segments, and then each segment is optimized either using compiler-based data distributions (at compile-time) or using dynamic migration (at runtime). In selecting the optimization strategy to use for a program segment, we use a criterion based on the number of compile-time analyzable references in loops. To test the effectiveness of our strategy in optimizing memory locality of applications, we implemented it and compared its performance with that of several other techniques such as compiler-directed data distribution and OS-directed dynamic page migration. Our experimental results obtained through simulation indicate that our hybrid strategy outperforms other strategies and achieves the best performance for a set of codes with regular, irregular, and mixed (regular + irregular) access patterns.

Original languageEnglish (US)
Title of host publicationLanguages and Compilers for Parallel Computing - 15th Workshop, LCPC 2002, Revised Papers
Pages111-125
Number of pages15
DOIs
StatePublished - Dec 1 2005
Event15th Workshop on Languages and Compilers for Parallel Computing, LCPC 2002 - College Park, MD, United States
Duration: Jul 25 2002Jul 27 2002

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume2481 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other15th Workshop on Languages and Compilers for Parallel Computing, LCPC 2002
CountryUnited States
CityCollege Park, MD
Period7/25/027/27/02

Fingerprint

Data Distribution
Locality
Migration
Data storage equipment
Compiler
Irregular
Parallelism
Strategy
Software
Optimization
Evaluate
Experimental Results
Architecture
Simulation

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Kadayif, I., Kandemir, M., & Choudhary, A. (2005). A hybrid strategy based on data distribution and migration for optimizing memory locality. In Languages and Compilers for Parallel Computing - 15th Workshop, LCPC 2002, Revised Papers (pp. 111-125). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 2481 LNCS). https://doi.org/10.1007/11596110_8
Kadayif, I. ; Kandemir, Mahmut ; Choudhary, A. / A hybrid strategy based on data distribution and migration for optimizing memory locality. Languages and Compilers for Parallel Computing - 15th Workshop, LCPC 2002, Revised Papers. 2005. pp. 111-125 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{4a5a32a6265846e88ab36f38cd8dabea,
title = "A hybrid strategy based on data distribution and migration for optimizing memory locality",
abstract = "The performance of a NUMA architecture depends on the efficient use of local memory. Therefore, software-level techniques that improve memory locality (in addition to parallelism) are extremely important to extract the best performance from these architectures. The proposed solutions so far include OS-based automatic data migrations and compiler-based static/dynamic data distributions. This paper proposes and evaluates a hybrid strategy for optimizing memory locality in NUMA architectures. In this strategy, we employ both compiler-directed data distribution and OS-directed dynamic page migration. More specifically, a given program code is first divided into segments, and then each segment is optimized either using compiler-based data distributions (at compile-time) or using dynamic migration (at runtime). In selecting the optimization strategy to use for a program segment, we use a criterion based on the number of compile-time analyzable references in loops. To test the effectiveness of our strategy in optimizing memory locality of applications, we implemented it and compared its performance with that of several other techniques such as compiler-directed data distribution and OS-directed dynamic page migration. Our experimental results obtained through simulation indicate that our hybrid strategy outperforms other strategies and achieves the best performance for a set of codes with regular, irregular, and mixed (regular + irregular) access patterns.",
author = "I. Kadayif and Mahmut Kandemir and A. Choudhary",
year = "2005",
month = "12",
day = "1",
doi = "10.1007/11596110_8",
language = "English (US)",
isbn = "3540307818",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
pages = "111--125",
booktitle = "Languages and Compilers for Parallel Computing - 15th Workshop, LCPC 2002, Revised Papers",

}

Kadayif, I, Kandemir, M & Choudhary, A 2005, A hybrid strategy based on data distribution and migration for optimizing memory locality. in Languages and Compilers for Parallel Computing - 15th Workshop, LCPC 2002, Revised Papers. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 2481 LNCS, pp. 111-125, 15th Workshop on Languages and Compilers for Parallel Computing, LCPC 2002, College Park, MD, United States, 7/25/02. https://doi.org/10.1007/11596110_8

A hybrid strategy based on data distribution and migration for optimizing memory locality. / Kadayif, I.; Kandemir, Mahmut; Choudhary, A.

Languages and Compilers for Parallel Computing - 15th Workshop, LCPC 2002, Revised Papers. 2005. p. 111-125 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 2481 LNCS).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - A hybrid strategy based on data distribution and migration for optimizing memory locality

AU - Kadayif, I.

AU - Kandemir, Mahmut

AU - Choudhary, A.

PY - 2005/12/1

Y1 - 2005/12/1

N2 - The performance of a NUMA architecture depends on the efficient use of local memory. Therefore, software-level techniques that improve memory locality (in addition to parallelism) are extremely important to extract the best performance from these architectures. The proposed solutions so far include OS-based automatic data migrations and compiler-based static/dynamic data distributions. This paper proposes and evaluates a hybrid strategy for optimizing memory locality in NUMA architectures. In this strategy, we employ both compiler-directed data distribution and OS-directed dynamic page migration. More specifically, a given program code is first divided into segments, and then each segment is optimized either using compiler-based data distributions (at compile-time) or using dynamic migration (at runtime). In selecting the optimization strategy to use for a program segment, we use a criterion based on the number of compile-time analyzable references in loops. To test the effectiveness of our strategy in optimizing memory locality of applications, we implemented it and compared its performance with that of several other techniques such as compiler-directed data distribution and OS-directed dynamic page migration. Our experimental results obtained through simulation indicate that our hybrid strategy outperforms other strategies and achieves the best performance for a set of codes with regular, irregular, and mixed (regular + irregular) access patterns.

AB - The performance of a NUMA architecture depends on the efficient use of local memory. Therefore, software-level techniques that improve memory locality (in addition to parallelism) are extremely important to extract the best performance from these architectures. The proposed solutions so far include OS-based automatic data migrations and compiler-based static/dynamic data distributions. This paper proposes and evaluates a hybrid strategy for optimizing memory locality in NUMA architectures. In this strategy, we employ both compiler-directed data distribution and OS-directed dynamic page migration. More specifically, a given program code is first divided into segments, and then each segment is optimized either using compiler-based data distributions (at compile-time) or using dynamic migration (at runtime). In selecting the optimization strategy to use for a program segment, we use a criterion based on the number of compile-time analyzable references in loops. To test the effectiveness of our strategy in optimizing memory locality of applications, we implemented it and compared its performance with that of several other techniques such as compiler-directed data distribution and OS-directed dynamic page migration. Our experimental results obtained through simulation indicate that our hybrid strategy outperforms other strategies and achieves the best performance for a set of codes with regular, irregular, and mixed (regular + irregular) access patterns.

UR - http://www.scopus.com/inward/record.url?scp=33745137783&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=33745137783&partnerID=8YFLogxK

U2 - 10.1007/11596110_8

DO - 10.1007/11596110_8

M3 - Conference contribution

AN - SCOPUS:33745137783

SN - 3540307818

SN - 9783540307815

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 111

EP - 125

BT - Languages and Compilers for Parallel Computing - 15th Workshop, LCPC 2002, Revised Papers

ER -

Kadayif I, Kandemir M, Choudhary A. A hybrid strategy based on data distribution and migration for optimizing memory locality. In Languages and Compilers for Parallel Computing - 15th Workshop, LCPC 2002, Revised Papers. 2005. p. 111-125. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/11596110_8