Pre-prints

Permanent link for this collectionhttps://hdl.handle.net/2022/13011

Browse

Now showing 1 - 4 of 4

Performance Characteristics of Virtualized GPUs for Deep Learning
(2019-10) Michael, Scott; Teige, Scott; Li, Junjie; Lowe, John Michael; Turner, George; Henschel, Robert
As deep learning techniques and algorithms become more and more common in scientific workflows, HPC centers are grappling with how best to provide GPU resources and support deep learning workloads. One novel method of deployment is to virtualize GPU resources allowing for multiple VM instances to have logically distinct virtual GPUs (vGPUs) on a shared physical GPU. However, there are many operational and performance implications to consider before deploying a vGPU service in an HPC center. In this paper, we investigate the performance characteristics of vGPUs for both traditional HPC workloads and for deep learning training and inference workloads. Using NVIDIA’s vDWS virtualization software, we perform a series of HPC and deep learning benchmarks on both non-virtualized (bare metal) and vGPUs of various sizes and configurations. We report on several of the challenges we discovered in deploying and operating a variety of virtualized instance sizes and configurations. We find that the overhead of virtualization on HPC workloads is generally < 10%, and can vary considerably for deep learning, depending on the task.
Scalable Quality Assurance for Neuroimaging (SQAN): Quality Control for Medical Imaging Project
(2019) Gopu, A; Young, M.D.; West, J.; Parasivam, M.; Avena-Koenigsberger, A.
Medical imaging, a key component in the clinical diagnosis of and research on neurodegenerative diseases such as Alzheimer’s and Parkinson’s, and various brain cancers, can generate massive datasets in the order of tens to even hundreds of thousands of images for a single subject over the course of a study, and studies typically include hundreds of subjects. Quality assurance (QA) plays a critical role in guaranteeing high reliability of medical imaging research ; it requires continuous involvement by all stakeholders and use of specific quality control (QC) methods. Imaging is very costly, and many projects lack funds to reacquire images if QC issues are detected later, or this reacquisition may be impractical based on the distance between the patient and the facility, or difficulties inherent in the illness. Therefore, appropriate QC methods are imperative, as they can rapidly identify bad data and areas likely to require to post-processing correction or real-time re-acquisition. While many useful QC methods including automated or semi-automated procedures exist, they are often complex, limited in informational scope, require time consuming manual techniques, and lack documentation; they are generally designed for specific use-cases, making integration with other setups difficult. In this paper, we summarize the work done by the the Indiana University Scalable Compute Archive (IU SCA) team and the RADY Imaging Center (Indianapolis) between 2015-now on the Scalable Quality Assurance for Neuroimaging project (SQAN - pronounced “scan”) software suite. SQAN runs a comprehensive QC pipeline, comparing various facets of each scan for each subject with an appropriate template. It ensures all the scans required across each modality expected by a research study’s pre-defined protocol are present with the expected image counts; and that values match exactly, or within a percentage threshold of the template value. It safely ignores keywords expected to differ between the template and subject exam (e.g. timestamps, subject demographics). Outside of coding, our project required a significant amount of engagement and feedback cycles with researchers, scanner technologists, and data scientists, each of whom approach QC with their own unique priorities and insights. Since Fall 2017, a fully operational production SQAN service instance has supported the Indianapolis-based RADY Imaging Center (RADY-SQAN) providing QC for 50+ research projects; it has QC’ed nearly 3 million images and validated over 600 million metadata tags. A more detailed description of the QC capabilities is available in an accompanying technical description paper.
Scalable Quality Assurance for Neuroimaging (SQAN) - Technical design including software application stack
(2019) Young, M.D.; Avena-Koenigsberger, A.; Hayashi, S.; Gopu, A.; West, J.; Paramasivam, M.; Perigo, R.
In medical research a single magnetic resonance imaging (MRI) exam of a single subject can produce hundreds of thousands of individual images and millions of key-value metadata pairs which must be verified to ensure instrument performance and compliance with the research protocol. Here we describe a system to address this concern, the Scalable Quality Assurance for Neuroimaging (SQAN), an open-source suite of tools used to extract metadata and perform quality control (QC) protocol and instrumental validation on medical imaging files (e.g. DICOM). The design features several discrete components, including: systems for receiving and storing incoming live data from remote imaging centers; processes for performing quality control validation on new and archive data; an Application Programming Interface (API) for mediating secure authorized access to imaging data and QC results; and a web-based User Interface (UI) for viewing stored data, QC results, modifying QC templates and access controls, commenting on QC issues, and alerting affected researchers, and re-running QC tests as needed. This paper is the second in a series, with the first discussing the background, motivations, and broad overview of SQAN as a project. In this paper we will provide a low-level technical description of the systems, methods, and infrastructure of the SQAN application stack. In addition to a further examination of the principal SQAN components we will explore additional features, including: anonymization of electronically Protected Health Information (ePHI); secure data transfer from remote imaging centers; extraction and compression of imaging metadata; optimized mongo database structure; and the QC templates and validations, including exclusions and handling of edge-cases, which are numerous. We will also describe the lifecycle of typical medical imaging exam, from acquisition through QC acceptance.
Assessment of financial returns on investments in cyberinfrastructure facilities: A survey of current methods
(2019) Stewart, Craig A.; Hancock, David Y.; Wernert, Julie; Furlani, Thomas; Lifka, David; Sill, Alan; Berente, Nicholas; McMullen, Donald F.; Cheatham, Thomas; Apon, Amy; Payne, Ron; Slavin, Shawn D.
In recent years, considerable attention has been given to assessing the value of investments in cyberinfrastructure (CI). This paper includes a survey of current methods for the assessment of financial returns on investment (ROI) in CI. Applying the financial concept of ROI proves challenging with regard to a service that, in most academic environments, does not generate a “sold amount” such as one would find in the buying and selling of stocks. The paper concludes with a discussion of future research directions and challenges in the assessment of financial ROI in CI. This work is intended less as a definitive guide than as a starting point for further exploration in the assessment of CI’s value for scientific research.

Browse

Recent Submissions