Workload Characterization of Nondeterministic Programs Parallelized by STATS

Enrico A. Deiana, Simone Campanoni

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

Abstract

Chip Multiprocessors (CMP) are everywhere, from mobile systems, to servers. Thread Level Parallelism (TLP) is the characteristic of a program that makes use of the parallel cores of a CMP to generate performance. Despite all efforts for creating TLP, multiple cores are still underutilized even though we have been in the multicore era for more than a decade. Recently, a new approach called STATS has been proposed to generate additional TLP for complex and irregular nondeterministic programs. STATS allows a developer to describe application-specific information that its compiler uses to automatically generate a new source of TLP. This new source of TLP increases with the size of the input and it has the potential to generate scalable performance with the number of cores. Even though STATS obtains most of its potential, some of it is still unreached. This paper identifies and characterizes the sources of overhead that are currently blocking STATS parallelized programs to achieve their full potential. To this end, we characterized the workloads generated by the STATS compiler on a 28 core Intel-based machine (dual-socket). This paper shows that the performance loss is due to a combination of factors: some can be optimized via engineering efforts and some require a deeper evolution of STATS. We also highlight potential solutions to significantly reduce most of this overhead. Exploiting these insights will unblock scalable performance for the parallel binaries generated by STATS.

Original languageEnglish (US)
Title of host publicationProceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages190-201
Number of pages12
ISBN (Electronic)9781728107462
DOIs
StatePublished - Apr 22 2019
Event2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019 - Madison, United States
Duration: Mar 24 2019Mar 26 2019

Publication series

NameProceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019

Conference

Conference2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019
CountryUnited States
CityMadison
Period3/24/193/26/19

Fingerprint

Servers

Keywords

  • Nondeterminism
  • Parallelizing compiler
  • Speculation
  • Workload characterization

ASJC Scopus subject areas

  • Hardware and Architecture
  • Software
  • Safety, Risk, Reliability and Quality

Cite this

Deiana, E. A., & Campanoni, S. (2019). Workload Characterization of Nondeterministic Programs Parallelized by STATS. In Proceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019 (pp. 190-201). [8695670] (Proceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ISPASS.2019.00032
Deiana, Enrico A. ; Campanoni, Simone. / Workload Characterization of Nondeterministic Programs Parallelized by STATS. Proceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019. Institute of Electrical and Electronics Engineers Inc., 2019. pp. 190-201 (Proceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019).
@inproceedings{e231caba417d4ae2851bc1087bc21a3c,
title = "Workload Characterization of Nondeterministic Programs Parallelized by STATS",
abstract = "Chip Multiprocessors (CMP) are everywhere, from mobile systems, to servers. Thread Level Parallelism (TLP) is the characteristic of a program that makes use of the parallel cores of a CMP to generate performance. Despite all efforts for creating TLP, multiple cores are still underutilized even though we have been in the multicore era for more than a decade. Recently, a new approach called STATS has been proposed to generate additional TLP for complex and irregular nondeterministic programs. STATS allows a developer to describe application-specific information that its compiler uses to automatically generate a new source of TLP. This new source of TLP increases with the size of the input and it has the potential to generate scalable performance with the number of cores. Even though STATS obtains most of its potential, some of it is still unreached. This paper identifies and characterizes the sources of overhead that are currently blocking STATS parallelized programs to achieve their full potential. To this end, we characterized the workloads generated by the STATS compiler on a 28 core Intel-based machine (dual-socket). This paper shows that the performance loss is due to a combination of factors: some can be optimized via engineering efforts and some require a deeper evolution of STATS. We also highlight potential solutions to significantly reduce most of this overhead. Exploiting these insights will unblock scalable performance for the parallel binaries generated by STATS.",
keywords = "Nondeterminism, Parallelizing compiler, Speculation, Workload characterization",
author = "Deiana, {Enrico A.} and Simone Campanoni",
year = "2019",
month = "4",
day = "22",
doi = "10.1109/ISPASS.2019.00032",
language = "English (US)",
series = "Proceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
pages = "190--201",
booktitle = "Proceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019",
address = "United States",

}

Deiana, EA & Campanoni, S 2019, Workload Characterization of Nondeterministic Programs Parallelized by STATS. in Proceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019., 8695670, Proceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019, Institute of Electrical and Electronics Engineers Inc., pp. 190-201, 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019, Madison, United States, 3/24/19. https://doi.org/10.1109/ISPASS.2019.00032

Workload Characterization of Nondeterministic Programs Parallelized by STATS. / Deiana, Enrico A.; Campanoni, Simone.

Proceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019. Institute of Electrical and Electronics Engineers Inc., 2019. p. 190-201 8695670 (Proceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019).

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

TY - GEN

T1 - Workload Characterization of Nondeterministic Programs Parallelized by STATS

AU - Deiana, Enrico A.

AU - Campanoni, Simone

PY - 2019/4/22

Y1 - 2019/4/22

N2 - Chip Multiprocessors (CMP) are everywhere, from mobile systems, to servers. Thread Level Parallelism (TLP) is the characteristic of a program that makes use of the parallel cores of a CMP to generate performance. Despite all efforts for creating TLP, multiple cores are still underutilized even though we have been in the multicore era for more than a decade. Recently, a new approach called STATS has been proposed to generate additional TLP for complex and irregular nondeterministic programs. STATS allows a developer to describe application-specific information that its compiler uses to automatically generate a new source of TLP. This new source of TLP increases with the size of the input and it has the potential to generate scalable performance with the number of cores. Even though STATS obtains most of its potential, some of it is still unreached. This paper identifies and characterizes the sources of overhead that are currently blocking STATS parallelized programs to achieve their full potential. To this end, we characterized the workloads generated by the STATS compiler on a 28 core Intel-based machine (dual-socket). This paper shows that the performance loss is due to a combination of factors: some can be optimized via engineering efforts and some require a deeper evolution of STATS. We also highlight potential solutions to significantly reduce most of this overhead. Exploiting these insights will unblock scalable performance for the parallel binaries generated by STATS.

AB - Chip Multiprocessors (CMP) are everywhere, from mobile systems, to servers. Thread Level Parallelism (TLP) is the characteristic of a program that makes use of the parallel cores of a CMP to generate performance. Despite all efforts for creating TLP, multiple cores are still underutilized even though we have been in the multicore era for more than a decade. Recently, a new approach called STATS has been proposed to generate additional TLP for complex and irregular nondeterministic programs. STATS allows a developer to describe application-specific information that its compiler uses to automatically generate a new source of TLP. This new source of TLP increases with the size of the input and it has the potential to generate scalable performance with the number of cores. Even though STATS obtains most of its potential, some of it is still unreached. This paper identifies and characterizes the sources of overhead that are currently blocking STATS parallelized programs to achieve their full potential. To this end, we characterized the workloads generated by the STATS compiler on a 28 core Intel-based machine (dual-socket). This paper shows that the performance loss is due to a combination of factors: some can be optimized via engineering efforts and some require a deeper evolution of STATS. We also highlight potential solutions to significantly reduce most of this overhead. Exploiting these insights will unblock scalable performance for the parallel binaries generated by STATS.

KW - Nondeterminism

KW - Parallelizing compiler

KW - Speculation

KW - Workload characterization

UR - http://www.scopus.com/inward/record.url?scp=85065424848&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85065424848&partnerID=8YFLogxK

U2 - 10.1109/ISPASS.2019.00032

DO - 10.1109/ISPASS.2019.00032

M3 - Conference contribution

T3 - Proceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019

SP - 190

EP - 201

BT - Proceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019

PB - Institute of Electrical and Electronics Engineers Inc.

ER -

Deiana EA, Campanoni S. Workload Characterization of Nondeterministic Programs Parallelized by STATS. In Proceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019. Institute of Electrical and Electronics Engineers Inc. 2019. p. 190-201. 8695670. (Proceedings - 2019 IEEE International Symposium on Performance Analysis of Systems and Software, ISPASS 2019). https://doi.org/10.1109/ISPASS.2019.00032