Resource Estimation in High Performance Medical Image Computing
- PDF / 3,118,586 Bytes
- 11 Pages / 595.276 x 790.866 pts Page_size
- 72 Downloads / 165 Views
ORIGINAL ARTICLE
Resource Estimation in High Performance Medical Image Computing Rueben Banalagay & Kelsie Jade Covington & D.M. Wilkes & Bennett A. Landman
# Springer Science+Business Media New York 2014
Abstract Medical imaging analysis processes often involve the concatenation of many steps (e.g., multi-stage scripts) to integrate and realize advancements from image acquisition, image processing, and computational analysis. With the dramatic increase in data size for medical imaging studies (e.g., improved resolution, higher throughput acquisition, shared databases), interesting study designs are becoming intractable or impractical on individual workstations and servers. Modern pipeline environments provide control structures to distribute computational load in high performance computing (HPC) environments. However, high performance computing environments are often shared resources, and scheduling computation across these resources necessitates higher level modeling of resource utilization. Submission of ‘jobs’ requires an estimate of the CPU runtime and memory usage. The resource requirements for medical image processing algorithms are difficult to predict since the requirements can vary greatly between different machines, different execution instances, and different data inputs. Poor resource estimates can lead to wasted resources in high performance environments due to incomplete executions and extended queue wait times. Hence, resource estimation is becoming a major hurdle for medical image processing algorithms to efficiently leverage high performance computing environments. Herein, we present our implementation of a resource estimation system to overcome these difficulties and ultimately provide users with the ability to more efficiently utilize high performance computing resources.
R. Banalagay : K. J. Covington : D. Wilkes : B. A. Landman Electrical Engineering, Vanderbilt University EECS, 2301 Vandervilt P1, PO Box 351679 Station B, Nashville 37235-1679, TN, USA B. A. Landman (*) Biomedical Engineering, Vanderbilt University, Nashville 37235, TN, USA e-mail: [email protected]
Keywords Java image science toolkit . JIST RRID:nlx_151344 . Resource estimation . High performance computing . Decision trees
Introduction With the increased use of more complex algorithms on progressively larger amounts of medical image data, there is a growing need to use high performance computing environments, such as grid clusters, to process results quickly and efficiently (e.g., (Rex et al. 2003; Pieper et al. 2006; Sheehan et al. 1996; Parker and Johnson 1995; Lucas et al. 1992; Konstantinides and Rasure 1994; Lucas et al. 2010)). However, high performance computing (HPC) environments are often shared resources, and submitting software tasks requires an estimate of the CPU runtime and memory usage needed for execution. Moreover, higher level data management systems (e.g., XNAT database, pipeline, and web portal (Marcus et al. 2005)) are increasingly being used to manage both archival and pipeline process
Data Loading...
