SparTen: Leveraging Kokkos for On-node Parallelism in a Second-Order Method for Fitting Canonical Polyadic Tensor Models to Poisson Data

Keita Teranishi; Daniel M. Dunlavy; Jeremy M. Myers; Richard F. Barrett

doi:10.1109/HPEC43674.2020.9286251

SparTen: Leveraging Kokkos for On-node Parallelism in a Second-Order Method for Fitting Canonical Polyadic Tensor Models to Poisson Data

Keita Teranishi
, Daniel M. Dunlavy
, Jeremy M. Myers
, Richard F. Barrett

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

9 Scopus citations

Abstract

Canonical Polyadic tensor decomposition using alternate Poisson regression (CP-APR) is an effective analysis tool for large sparse count datasets. One of the variants using projected damped Newton optimization for row subproblems (PDNR) offers quadratic convergence and is amenable to parallelization. Despite its potential effectiveness, PDNR performance on modern high performance computing (HPC) systems is not well understood. To remedy this, we have developed a parallel implementation of PDNR using Kokkos, a performance portable parallel programming framework supporting efficient runtime of a single code base on multiple HPC systems. We demonstrate that the performance of parallel PDNR can be poor if load imbalance associated with the irregular distribution of nonzero entries in the tensor data is not addressed. Preliminary results using tensors from the FROSTT data set indicate that using multiple kernels to address this imbalance when solving the PDNR row subproblems in parallel can improve performance, with up to 80% speedup on CPUs and 10-fold speedup on NVIDIA GPUs.

Original language	English
Title of host publication	2020 IEEE High Performance Extreme Computing Conference, HPEC 2020
Publisher	Institute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)	9781728192192
DOIs	https://doi.org/10.1109/HPEC43674.2020.9286251
State	Published - Sep 22 2020
Externally published	Yes
Event	2020 IEEE High Performance Extreme Computing Conference, HPEC 2020 - Virtual, Waltham, United States Duration: Sep 21 2020 → Sep 25 2020

Publication series

Name	2020 IEEE High Performance Extreme Computing Conference, HPEC 2020

Conference

Conference	2020 IEEE High Performance Extreme Computing Conference, HPEC 2020
Country/Territory	United States
City	Virtual, Waltham
Period	09/21/20 → 09/25/20

Funding

Sandia National Laboratories is a multimission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Hon-eywell International Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525.

Keywords

GPU
Kokkos
Poisson factorization
multicore
tensor decomposition

Access to Document

10.1109/HPEC43674.2020.9286251

Cite this

Teranishi, K., Dunlavy, D. M., Myers, J. M., & Barrett, R. F. (2020). SparTen: Leveraging Kokkos for On-node Parallelism in a Second-Order Method for Fitting Canonical Polyadic Tensor Models to Poisson Data. In 2020 IEEE High Performance Extreme Computing Conference, HPEC 2020 Article 9286251 (2020 IEEE High Performance Extreme Computing Conference, HPEC 2020). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/HPEC43674.2020.9286251

Teranishi, Keita ; Dunlavy, Daniel M. ; Myers, Jeremy M. et al. / SparTen : Leveraging Kokkos for On-node Parallelism in a Second-Order Method for Fitting Canonical Polyadic Tensor Models to Poisson Data. 2020 IEEE High Performance Extreme Computing Conference, HPEC 2020. Institute of Electrical and Electronics Engineers Inc., 2020. (2020 IEEE High Performance Extreme Computing Conference, HPEC 2020).

@inproceedings{5f105f8c78674f6da4b8369ca5b36adc,

title = "SparTen: Leveraging Kokkos for On-node Parallelism in a Second-Order Method for Fitting Canonical Polyadic Tensor Models to Poisson Data",

abstract = "Canonical Polyadic tensor decomposition using alternate Poisson regression (CP-APR) is an effective analysis tool for large sparse count datasets. One of the variants using projected damped Newton optimization for row subproblems (PDNR) offers quadratic convergence and is amenable to parallelization. Despite its potential effectiveness, PDNR performance on modern high performance computing (HPC) systems is not well understood. To remedy this, we have developed a parallel implementation of PDNR using Kokkos, a performance portable parallel programming framework supporting efficient runtime of a single code base on multiple HPC systems. We demonstrate that the performance of parallel PDNR can be poor if load imbalance associated with the irregular distribution of nonzero entries in the tensor data is not addressed. Preliminary results using tensors from the FROSTT data set indicate that using multiple kernels to address this imbalance when solving the PDNR row subproblems in parallel can improve performance, with up to 80\% speedup on CPUs and 10-fold speedup on NVIDIA GPUs.",

keywords = "GPU, Kokkos, Poisson factorization, multicore, tensor decomposition",

author = "Keita Teranishi and Dunlavy, \{Daniel M.\} and Myers, \{Jeremy M.\} and Barrett, \{Richard F.\}",

note = "Publisher Copyright: {\textcopyright} 2020 IEEE.; 2020 IEEE High Performance Extreme Computing Conference, HPEC 2020 ; Conference date: 21-09-2020 Through 25-09-2020",

year = "2020",

month = sep,

day = "22",

doi = "10.1109/HPEC43674.2020.9286251",

language = "English",

series = "2020 IEEE High Performance Extreme Computing Conference, HPEC 2020",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

booktitle = "2020 IEEE High Performance Extreme Computing Conference, HPEC 2020",

}

Teranishi, K, Dunlavy, DM, Myers, JM & Barrett, RF 2020, SparTen: Leveraging Kokkos for On-node Parallelism in a Second-Order Method for Fitting Canonical Polyadic Tensor Models to Poisson Data. in 2020 IEEE High Performance Extreme Computing Conference, HPEC 2020., 9286251, 2020 IEEE High Performance Extreme Computing Conference, HPEC 2020, Institute of Electrical and Electronics Engineers Inc., 2020 IEEE High Performance Extreme Computing Conference, HPEC 2020, Virtual, Waltham, United States, 09/21/20. https://doi.org/10.1109/HPEC43674.2020.9286251

SparTen: Leveraging Kokkos for On-node Parallelism in a Second-Order Method for Fitting Canonical Polyadic Tensor Models to Poisson Data. / Teranishi, Keita; Dunlavy, Daniel M.; Myers, Jeremy M. et al.
2020 IEEE High Performance Extreme Computing Conference, HPEC 2020. Institute of Electrical and Electronics Engineers Inc., 2020. 9286251 (2020 IEEE High Performance Extreme Computing Conference, HPEC 2020).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - SparTen

T2 - 2020 IEEE High Performance Extreme Computing Conference, HPEC 2020

AU - Teranishi, Keita

AU - Dunlavy, Daniel M.

AU - Myers, Jeremy M.

AU - Barrett, Richard F.

PY - 2020/9/22

Y1 - 2020/9/22

N2 - Canonical Polyadic tensor decomposition using alternate Poisson regression (CP-APR) is an effective analysis tool for large sparse count datasets. One of the variants using projected damped Newton optimization for row subproblems (PDNR) offers quadratic convergence and is amenable to parallelization. Despite its potential effectiveness, PDNR performance on modern high performance computing (HPC) systems is not well understood. To remedy this, we have developed a parallel implementation of PDNR using Kokkos, a performance portable parallel programming framework supporting efficient runtime of a single code base on multiple HPC systems. We demonstrate that the performance of parallel PDNR can be poor if load imbalance associated with the irregular distribution of nonzero entries in the tensor data is not addressed. Preliminary results using tensors from the FROSTT data set indicate that using multiple kernels to address this imbalance when solving the PDNR row subproblems in parallel can improve performance, with up to 80% speedup on CPUs and 10-fold speedup on NVIDIA GPUs.

AB - Canonical Polyadic tensor decomposition using alternate Poisson regression (CP-APR) is an effective analysis tool for large sparse count datasets. One of the variants using projected damped Newton optimization for row subproblems (PDNR) offers quadratic convergence and is amenable to parallelization. Despite its potential effectiveness, PDNR performance on modern high performance computing (HPC) systems is not well understood. To remedy this, we have developed a parallel implementation of PDNR using Kokkos, a performance portable parallel programming framework supporting efficient runtime of a single code base on multiple HPC systems. We demonstrate that the performance of parallel PDNR can be poor if load imbalance associated with the irregular distribution of nonzero entries in the tensor data is not addressed. Preliminary results using tensors from the FROSTT data set indicate that using multiple kernels to address this imbalance when solving the PDNR row subproblems in parallel can improve performance, with up to 80% speedup on CPUs and 10-fold speedup on NVIDIA GPUs.

KW - GPU

KW - Kokkos

KW - Poisson factorization

KW - multicore

KW - tensor decomposition

UR - https://www.scopus.com/pages/publications/85099384524

U2 - 10.1109/HPEC43674.2020.9286251

DO - 10.1109/HPEC43674.2020.9286251

M3 - Conference contribution

AN - SCOPUS:85099384524

T3 - 2020 IEEE High Performance Extreme Computing Conference, HPEC 2020

BT - 2020 IEEE High Performance Extreme Computing Conference, HPEC 2020

PB - Institute of Electrical and Electronics Engineers Inc.

Y2 - 21 September 2020 through 25 September 2020

ER -

Teranishi K, Dunlavy DM, Myers JM, Barrett RF. SparTen: Leveraging Kokkos for On-node Parallelism in a Second-Order Method for Fitting Canonical Polyadic Tensor Models to Poisson Data. In 2020 IEEE High Performance Extreme Computing Conference, HPEC 2020. Institute of Electrical and Electronics Engineers Inc. 2020. 9286251. (2020 IEEE High Performance Extreme Computing Conference, HPEC 2020). doi: 10.1109/HPEC43674.2020.9286251

SparTen: Leveraging Kokkos for On-node Parallelism in a Second-Order Method for Fitting Canonical Polyadic Tensor Models to Poisson Data

Abstract

Publication series

Conference

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this