1
Return

PMIO: Optimizing Parallel Collective I/O Accesses through Node-Local Persistent Memory

delete2026-04-15
delete0
PRE
AI
K
Keegan Sanchez
R
Rubayet Rongon
A
Alex Gavin
S
Suren Byna
K
Kesheng Wu
X
Xuechen Zhang *
DOI：10.1016/j.jpdc.2026.105275delete
deleteOriginal
deleteOriginal request for help
deleteShare
deleteSave
Abstract

Abstract

En 中文
• We identify the major bottleneck of MPI Collective I/O on supercomputers. • We design PMIO to increase the size of collective I/O buffers leveraging node-local persistent memory. • Our experimental results show that PMIO outperforms ROMIO collective I/Os by up to 151x on supercomputers.
Keywords:
MPI Collective I/O
Persistent Memory
Supercomputers
I/O Optimization
Parallel Computing

Journal

Journal of Parallel and Distributed Computing cover
Journal of Parallel and Distributed Computing
IF:
4
Papers:
3.8K
Citations:
4.8K

Organization

K
kennesaw state university
Scholars:
81
Papers: 57
Citations: 0
L
Lawrence Berkeley National Laboratory
Scholars:
1.5W
Papers: 1.1W
Citations: 6.1W
O
Ohio State University
Scholars:
4.0W
Papers: 3.2W
Citations: 80
W
Washington State University
Scholars:
513
Papers: 248
Citations: 1
Cited Papers

Cited Papers

Citing Papers

Citing Papers