Hardware-accelerated high-resolution video coding in Virtual Network Functions

Abstract

Network Function Virtualization (NFV) has become a widely acclaimed approach to facilitate the management and orchestration of network services. However, after rapidly achieving a widespread success, NFV is now challenged by the overwhelming demand of computing power originated by the never-ending growth of innovative applications coming from the Internet world. To overcome this problem, the use of h/w acceleration combined with NFV has been proposed. This way, the computing performance of commodity servers can be greatly enhanced, without losing the advantages offered by NFV in service management. In this paper, to demonstrate the potentialities of NFV and h/w acceleration, a Virtual Network Function for video coding (video Transcoding Unit-vTU) is presented. The vTU is accelerated by a General Purpose GPU, and is based on Open Source software packages for media processing. The vTU architecture is firstly described in details. A thorough characterization of its computing performance is then reported, and the obtained results are compared to those achieved with non-accelerated and/or non-virtualized versions of the vTU itself. Also, the performance provided by an original, GPU accelerated version of the VP8 encoder is presented. The activities described in this paper have been carried out within the EU FP7 T-NOVA project.

Techno‐economic analysis of 5G immersive media services in cloud‐enabled small cell networks: The neutral host business model

“…The single transcoding rates μ 0 and μ 1 have been achieved with Nvidia M4000 GPUs. 20,24 The value N (D E , μ F, λ 0 ) = 110 has been obtained with D E equal to 5 seconds. However, if we increase D E to 10 seconds leaving μ F, λ 0 unchanged, N (D E , μ F, λ 0 ) becomes 310.…”

“…The reported parameters refer to a stadium with U TOTAL = 10 000 attending users; for the sake of simplicity, the number of silver users has been set to zero. The single transcoding rates μ 0 and μ 1 have been achieved with Nvidia M4000 GPUs . The value N (D E , μ F, λ 0 ) = 110 has been obtained with D E equal to 5 seconds.…”

“…*The validity of the EPS model has been investigated through a number of laboratory experiments (some of which were carried out during the SESAME project) summarized and discussed in other works. 20,24,26 The interested reader can specifically refer to the work of Paglierani et al 20 Moreover, one can use (5) with different values of (D E , μ F , λ 0 ), if the differentiated models are needed for gold and silver users. In this case, the minimum number M VPUOFF of VPU instances is…”

Deploying CPU-Intensive Applications on MEC in NFV Systems: The Immersive Video Use Case

“…However, most of the works available concerning MEC focus on traditional video networking functionalities, such as video caching or video streaming [5,[13][14][15][16][17][18][19][20][21][22][23][24]. Conversely, the problem of successfully combining end-user apps and ME apps to provide interactive and immersive video services to consumers on a peer-to-peer basis is a topic that has received little attention from researchers so far.…”

“…In this case, the use of High-Performance Computing (HPC) devices, such as Graphics Processing Unit(s) (GPUs), could provide significant benefits [25]. Nonetheless, though some results about the use of offloading techniques in virtualization infrastructure are available [22,25], the analysis of the computing performance achievable by ME systems when performing video-processing tasks has not been thoroughly analyzed yet.…”