Improving I/O Performance for Exascale Applications Through Online Data Layout Reorganization

Wan, Lipeng; Huebl, Axel; Gu, Jianhua; Poeschel, Franz; Gainaru, Ana; Wang, Ruonan; Chen, Jieyang; Liang, Xin; Dmitry, Ganyushin,; Munson, Todd; Foster, Ian; Vay, Jean‐Luc; Podhorszki, Norbert; Wu, Kesheng; Klasky, Scott

doi:10.1109/tpds.2021.3100784

Cited by 15 publications

(8 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Similarly, Mayes et al [21] (LUME) integrates single simulation modules with openPMD into a file-based loosely coupled pipeline for unified particle accelerator and lightsource modeling. Wan et al [24] investigate layout reorganization of parallel PIC simulation data to increase file IO performance, including downstream processing and data loads on finely chunked, distributed data. An established framework for loose coupling in the domain of Big Data is the JVM-based Apache Spark [26].…”

Section: Related Workmentioning

confidence: 99%

Transitioning from File-Based HPC Workflows to Streaming Data Pipelines with openPMD and ADIOS2

Poeschel

Juncheng

Godoy

et al. 2022

Driving Scientific and Engineering Discoveries Through the Integration of Experiment, Big Data, and Modeling and Simulation

Self Cite

View full text Add to dashboard Cite

This paper aims to create a transition path from file-based IO to streaming-based workflows for scientific applications in an HPC environment. By using the openPMP-api, traditional workflows limited by filesystem bottlenecks can be overcome and flexibly extended for in situ analysis. The openPMD-api is a library for the description of scientific data according to the Open Standard for Particle-Mesh Data (openPMD). Its approach towards recent challenges posed by hardware heterogeneity lies in the decoupling of data description in domain sciences, such as plasma physics simulations, from concrete implementations in hardware and IO. The streaming backend is provided by the ADIOS2 framework, developed at Oak Ridge National Laboratory. This paper surveys two openPMD-based loosely coupled setups to demonstrate flexible applicability and to evaluate performance. In loose coupling, as opposed to tight coupling, two (or more) applications are executed separately, e.g. in individual MPI contexts, yet cooperate by exchanging data. This way, a streaming-based workflow allows for standalone codes instead of tightlycoupled plugins, using a unified streaming-aware API and leveraging high-speed communication infrastructure available in modern compute clusters for massive data exchange. We determine new challenges in resource allocation and in the need of strategies for a flexible data distribution, demonstrating their influence on efficiency and scaling on the Summit compute system. The presented setups show the potential for a more flexible use of compute resources brought by streaming IO as well as the ability to increase throughput by avoiding filesystem bottlenecks.

show abstract

Section: Related Workmentioning

confidence: 99%

Transitioning from File-Based HPC Workflows to Streaming Data Pipelines with openPMD and ADIOS2

Poeschel

Juncheng

Godoy

et al. 2022

Driving Scientific and Engineering Discoveries Through the Integration of Experiment, Big Data, and Modeling and Simulation

Self Cite

View full text Add to dashboard Cite

show abstract

“…ADIOS's focus is to provide extreme-scale I/O capabilities on the world's largest supercomputers; it is used in science applications that generate data on the order of several petabytes. ADIOS is run in production in many HPC codes such as XGC [35], GENE [36], GEM [37], PIConGPU [38], WarpX [39], E3SM [40], LAMMPS [41], and others, providing over 1 Terabyte/second of I/O to the Summit GPFS file system [42] at ORNL.…”

Section: Adios Data Management Frameworkmentioning

confidence: 99%

Scalable training of graph convolutional neural networks for fast and accurate predictions of HOMO-LUMO gap in molecules

Choi

Zhang

Mehta

et al. 2022

Preprint

View full text Add to dashboard Cite

Graph Convolutional Neural Network (GCNN) is a popular class of deep learning (DL) models in material science to predict material properties from the graph representation of molecular structures. Training an accurate and comprehensive GCNN surrogate for molecular design requires large-scale graph datasets and is usually a time-consuming process. Recent advances in GPUs and distributed computing open a path to reduce the computational cost for GCNN training effectively. However, efficient utilization of high performance computing (HPC) resources for training requires simultaneously optimizing large-scale data management and scalable stochastic batched optimization techniques. In this work, we focus on building GCNN models on HPC systems to predict material properties of millions of molecules. We use HydraGNN, our in-house library for large-scale GCNN training, leveraging distributed data parallelism in PyTorch. We use ADIOS, a high-performance data management framework for efficient storage and reading of large molecular graph data. We perform parallel training on two open-source large-scale graph datasets to build a GCNN predictor for an important quantum property known as the HOMO-LUMO gap. We measure the scalability, accuracy, and convergence of our approach on two DOE supercomputers: the Summit supercomputer at the Oak Ridge Leadership Computing Facility (OLCF) and the Perlmutter system at the National Energy Research Scientific Computing Center (NERSC). We present our experimental results with HydraGNN showing i) reduction of data loading time up to 4.2 times compared with a conventional method and ii) linear scaling performance for training up to 1,024 GPUs on both Summit and Perlmutter.

show abstract

“…Many research efforts focus on improving the I/O infrastructure (the parallel file system or an I/O library, for instance) in a way that transparently benefits all applications running on a system [7]- [12]. Still, reaching peak I/O performance often still depends on application tuning [21], which is a more labor-intensive task because of the huge number of different applications that exist. That effort is justified for applications that are considered important because they are executed for long periods of time, or very often, hence its performance improvements result in better utilization of HPC resources [22].…”

Section: Related Workmentioning

confidence: 99%

I/O performance of multiscale finite element simulations on HPC environments

Boito

Gomes

Peyrondet

et al. 2022

2022 International Symposium on Computer Architecture and High Performance Computing Workshops (SBAC-PADW)

View full text Add to dashboard Cite

In this paper, we present MSLIO, a code to mimic the I/O behavior of multiscale simulations. Such an I/O kernel is useful for HPC research, as it can be executed more easily and more efficiently than the full simulations when researchers are interested in the I/O load only. We validate MSLIO by comparing it to the I/O performance of an actual simulation, and we then use it to test some possible improvements to the output routine of the MHM (Multiscale Hybrid Mixed) library.

show abstract

Improving I/O Performance for Exascale Applications Through Online Data Layout Reorganization

Cited by 15 publications

References 26 publications

Transitioning from File-Based HPC Workflows to Streaming Data Pipelines with openPMD and ADIOS2

Transitioning from File-Based HPC Workflows to Streaming Data Pipelines with openPMD and ADIOS2

Scalable training of graph convolutional neural networks for fast and accurate predictions of HOMO-LUMO gap in molecules

I/O performance of multiscale finite element simulations on HPC environments

Contact Info

Product

Resources

About