Application performance analysis and efficient execution on systems with multi-core CPUs, GPUs and MICs: a case study with microscopy image analysis

• Type of Material: Article
• Authors: George Teodoro, University of Brasília; Tahsin Kurc, Stony Brook University; Guilherme Andrade, Federal University of Minas Gerais; Jun Kong, Emory University; Renato Ferreira, Federal University of Minas Gerais; Joel Saltz, Stony Brook University
• Language: English
• Date: 2017-01-01
• Publisher: SAGE Publications
• Publication Version: Author Accepted Manuscript (After Peer Review)
• Copyright Statement: © 2017, © SAGE Publications
• Final Published Version (URL): http://dx.doi.org/10.1177/1094342015594519
• Title of Journal or Parent Work: International Journal of High Performance Computing Applications
• ISSN: 1094-3420
• Volume: 31
• Issue: 1
• Start Page: 32
• End Page: 51
• Grant/Funding Information: This work was supported in part by HHSN261200800001E and 1U24CA180924-01A1 from the NCI, R24HL085343 from the NHLBI, R01LM011119-01 and R01LM009239 from the NLM, RC4MD005964 from the NIH, National Institutes of Health (NIH) K25CA181503 and by CNPq, CAPES, FINEP, Fapemig, and INWEB.; This research used resources provided by the XSEDE Science Gateways program and the Keeneland Computing Facility at the Georgia Institute of Technology, which is supported by the NSF under Contract OCI-0910735.
• Keywords: Hybrid systems; MODEL; Computer Science; Computer Science, Interdisciplinary Applications; Computer Science, Theory & Methods; ALGORITHMS; cooperative execution; Intel Xeon Phi; Computer Science, Hardware & Architecture; ARCHITECTURES; image analysis; Technology; Science & Technology; GPGPU
• Research Categories: Computer Science; Biology, Bioinformatics

Abstract

We carry out a comparative performance study of multi-core CPUs, GPUs and Intel Xeon Phi (Many Integrated Core-MIC) with a microscopy image analysis application. We experimentally evaluate the performance of computing devices on core operations of the application. We correlate the observed performance with the characteristics of computing devices and data access patterns, computation complexities, and parallelization forms of the operations. The results show a significant variability in the performance of operations with respect to the device used. The performances of operations with regular data access are comparable or sometimes better on a MIC than that on a GPU. GPUs are more efficient than MICs for operations that access data irregularly, because of the lower bandwidth of the MIC for random data accesses. We propose new performance-aware scheduling strategies that consider variabilities in operation speedups. Our scheduling strategies significantly improve application performance compared to classic strategies in hybrid configurations.