The study presented here focuses on performance characteristics and trade-offs associated with running machine-learning tasks in multi-GPU environments on both on-site cloud computing resources and commercial cloud services (Azure). Specifically, this study examines these tradeoffs by examining the performance of training and fine-tuning of transformer-based deep-learning (DL) networks on clinical notes and data, a task of critical importance in the medical domain. To this end, we perform DL-related experiments on the widely deployed NVIDIA V100 GPUs and on the newer A100 GPUs connected via NVLink or PCIe. This study analyzes the execution time of major operations to train DL models and investigate popular options to optimize each of them. We examine and present the findings on the impacts that various operations (e.g. data loading into GPUs, training, fine-tuning), optimizations, and system configurations (single vs. multi-GPU, NVLink vs. PCIe) have on the overall training performance.

This study aims to effectively analyze and visualize the concept to concept network derived from the COVID-19 Open Research Dataset (CORD-19) dataset, where we have more than 48,000 concepts with more than 300,000 relationships between concepts. In analyzing networks, we focus on finding relationship patterns between the coronavirus disease 2019 (COVID-19) concepts and other concepts. Given the node and edge datasets, we construct directional graphs and calculate all pair shortest paths based on multiple edge weight schemes. However, statistical metrics are not sufficient to identify specific relationships represented in the network. Therefore, we also propose a visual analytics approach to effectively understand the knowledge graph. Our highly interactive visual analytics allows users to effectively analyze the evolving graphs and (COVID-19) concept nodes and other nodes related to the COVID-19 nodes. We envision that this study will pave the path to develop strategies to provide more accurate and scalable predictive analysis on knowledge graphs related to CORD19 and other biomedical knowledge graphs.

Deep learning, through the use of neural networks, has demonstrated remarkable ability to automate many routine tasks when presented with sufficient data for training. The neural network architecture (e.g. number of layers, types of layers, connections between layers, etc.) plays a critical role in determining what, if anything, the neural network is able to learn from the training data. The trend for neural network architectures, especially those trained on ImageNet, has been to grow ever deeper and more complex. The result has been ever increasing accuracy on benchmark datasets with the cost of increased computational demands. In this paper we demonstrate that neural network architectures can be automatically generated, tailored for a specific application, with dual objectives: accuracy of prediction and speed of prediction. Using MENNDL- an HPC-enabled software stack for neural architecture search-we generate a neural network with comparable accuracy to state-of-the-art networks on a cancer pathology dataset that is also 16× faster at inference. The speedup in inference is necessary because of the volume and velocity of cancer pathology data; specifically, the previous state-of-the-art networks are too slow for individual researchers without access to HPC systems to keep pace with the rate of data generation. Our new model enables researchers with modest computational resources to analyze newly generated data faster than it is collected.

Linear algebraic operations such as matrix manipulations form the kernel of many machine learning and other crucial algorithms. Scaling up as well as scaling out such algorithms are key to supporting large scale data analysis that require efficient processing over millions of data samples. To this end, we present, ARION, a hardware acceleration based approach for scaling-up individual tasks of Spark, a popular data-parallel analytics platform. We support both linear algebraic operations of between two dense matrices, and between sparse and dense matrices in distributed environments. ARION provides a flexible control of acceleration according to matrix density, along with efficient scheduling based on runtime resource utilization. We demonstrate the benefit of our approach for general matrix multiplication operations over large matrices with up to four billion elements by using Gramian matrix computation that is commonly used in machine learning. Experiments show that our approach achieves more than 2× and 1.5× end-to-end performance speedups for dense and sparse matrices, respectively, and up to 57.04× faster computation compared to MLlib, a state of the art Spark-based implementation. This work is sponsored in part by the NSF under the grants: CNS-1565314, CNS-1405697, and CNS-1615411. The manuscript has been authored by UT-Battelle, LLC under Contract No. DE-AC05-00OR22725 with the U.S. Department of Energy. The United States Government retains and the publisher, by accepting the article for publication, acknowledges that the United States Government retains a non-exclusive, paid-up, irrevocable, world-wide license to publish or reproduce the published form of this manuscript, or allow others to do so, for United States Government purposes. The Department of Energy will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan (http://energy.gov/downloads/doe-public-access-plan). This research used resources of the Oak Ridge Leadership Computing Facility at the Oak Ridge National Laboratory, which is supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC05-00OR22725.

The construction of data analysis infrastructures that handle continuously accumulating data is quickly becoming an essential requirement for many organizations such as the U.S. Department of Energy (DOE). While DOE supports some of the largest computing facilities in the world, new analysis infrastructures like Apache Spark are difficult to implement. In this paper, we propose an on-demand Spark service that mitigates these difficulties, allowing facility users to flexibly create Spark instances quickly and easily. We define a systematic approach for creating these Spark instances and validate that optimal performance benefits are maintained. Using a series of benchmarks for algorithms that are commonly used in scientific workflows, we compared the behavior of Spark tasks using facility resources with that of an open research cloud that has a dedicated Spark infrastructure deployed. Finally, we leveraged a scientific use case from the Center of Nanophase Materials at the Oak Ridge National Laboratory to demonstrate the utility of using Spark in the computing facility.

In this paper, we pose and address some of the unique challenges in the analysis of scientific Big Data on supercomputing platforms. Our approach identifies, implements and scales numerical kernels that are critical to the instantiation of theory-inspired analytic workflows on modern computing architectures. We present the benefits of scalable kernels towards constructing algorithms such as principal component analysis and non-negative matrix factorization on an image-analysis use case at the Oak Ridge Leadership Computing Facility (OLCF). Based on experience with the use-case, we conclude that piecing scalable analytic kernels into user-defined analytic workflows are a flexible, modular and agile way to enable architecture-portable productivity for the data-intensive sciences.

Scaling-up scientific data analysis and machine learning algorithms for data-driven discovery is a grand challenge that we face today. Despite the growing need for analysis from science domains that are generating ‘Big Data’ from instruments and simulations, building high-performance analytical workflows of data-intensive algorithms have been daunting because: (i) the ‘Big Data’ hardware and software architecture landscape is constantly evolving, (ii) newer architectures impose new programming models, and (iii) data-parallel kernels of analysis algorithms and their performance facets on different architectures are poorly understood. To address these problems, we have: (i) identified scalable data-parallel kernels of popular data analysis algorithms, (ii) implemented ‘Mini-Apps’ of those kernels using different programming models (e.g. Map Reduce, MPI, etc.), (iii) benchmarked and validated the performance of the kernels in diverse architectures. In this paper, we discuss two of those Mini-Apps and show the execution of principal component analysis built as a workflow of the Mini-Apps. We show that Mini-Apps enable scientists to (i) write domain-specific data analysis code that scales on most HPC hardware and (ii) and offers the ability (most times with over a 10x speed-up) to analyze data sizes 100 times the size of what off-the-shelf desktop/workstations of today can handle.

Constellation's overarching goal is the federation of information from resources within an extreme-scale scientific collaboration to enable the scalable discovery of data and new knowledge pathways. The resource fabric is comprised of petascale supercomputers and storage systems, users, jobs, datasets and lifecycle artifacts. For an extreme-scale supercomputing center, normal operations can generate hundreds of millions of data products and metadata entries describing the resource fabric. Constellation federates the information extracted from the resources using a custom, transformative science graph network; constructs rich metadata indexes and higher-order derived metadata from the extracted information; and conducts scalable graph analytics to unravel hidden data pathways. Our implementation and deployment for a production, supercomputing facility shows that the graph can scale to more than 750 million vertices, its domain agnostic indexing can answer interesting science queries, and its analytics can aid in structural, topological and temporal analysis to identify usage hotspots.