Generic Virtualization Service (GVirtuS) is a new solution for enabling GPGPU on Virtual Machines or low powered devices. This paper focuses on the performance analysis that can be obtained using a GPGPU virtualized software. Recently, GVirtuS has been extended in order to support CUDA ancillary libraries with good results. Here, our aim is to analyze the applicability of this powerful tool to a real problem, which uses the NVIDIA cuFFT library. As case study we consider a simple denoising algorithm, implementing a virtualized GPU-parallel software based on the convolution theorem in order to perform the noise removal procedure in the frequency domain. We report some preliminary tests in both physical and virtualized environments to study and analyze the potential scalability of such an algorithm.
A virtualized software based on the NVIDIA cuFFT library for image denoising: performance analysis / Galletti, Ardelio; Marcellino, Livia; Montella, Raffaele; Santopietro, Vincenzo; Kosta, Sokol. - 113:(2017), pp. 496-501. (Intervento presentato al convegno 8th International Conference on Emerging Ubiquitous Systems and Pervasive Networks, EUSPN 2017 and the 7th International Conference on Current and Future Trends of Information and Communication Technologies in Healthcare, ICTH 2017 tenutosi a Lund; Sweden nel 2017) [10.1016/j.procs.2017.08.310].
A virtualized software based on the NVIDIA cuFFT library for image denoising: performance analysis
Kosta, Sokol
2017
Abstract
Generic Virtualization Service (GVirtuS) is a new solution for enabling GPGPU on Virtual Machines or low powered devices. This paper focuses on the performance analysis that can be obtained using a GPGPU virtualized software. Recently, GVirtuS has been extended in order to support CUDA ancillary libraries with good results. Here, our aim is to analyze the applicability of this powerful tool to a real problem, which uses the NVIDIA cuFFT library. As case study we consider a simple denoising algorithm, implementing a virtualized GPU-parallel software based on the convolution theorem in order to perform the noise removal procedure in the frequency domain. We report some preliminary tests in both physical and virtualized environments to study and analyze the potential scalability of such an algorithm.| File | Dimensione | Formato | |
|---|---|---|---|
|
Galletti_Virtualized_2017.pdf
accesso aperto
Tipologia:
Versione editoriale (versione pubblicata con il layout dell'editore)
Licenza:
Creative commons
Dimensione
383.08 kB
Formato
Adobe PDF
|
383.08 kB | Adobe PDF |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
