Empirical Analysis of Multicore CPU and GPU-Based Parallel Solutions to Sustain Throughput Needed by Scalable Proxy Servers for Protected Videos

• Leandro A. S. Gomes Universidade Federal do Pampa
• Bruno S. Neves Universidade Federal do Pampa
• Leonardo B. Pinho Universidade Federal do Pampa

Resumo

Proxy servers of scalable video distribution systems must perform not only efficient memory management but also adopt video protection mechanisms. This work proposes an adaptive mix-grained parallelization of AES ciphering algorithm to provide customized video segments to concurrent clients implemented using CUDA, Pthreads, and OpenMP in order to exploit multicore CPU or GPU. An evaluation is conducted using a server with Hyper-Threading (HT) capable multicore CPU and state-of-the-art 448-core GPU in contrast to a desktop using multicore CPU without HT and low-cost 128-core GPU. Overall, PCI-Express impacts CUDA's achievable throughput and HT affects the amount of cores and cooperative threads needed by application for Pthreads and even more for OpenMP. As expected, CUDA reaches higher throughput but a comparison between GPUs regarding thread occupancy demonstrates that greater core availability doesn't guarantee highest throughput, which will be demanded as network capacity of proxy servers is migrating from 1 to 10 Gbps.

Palavras-chave: Graphics processing units, Kernel, Multicore processing, Instruction sets, Throughput, Servers, Videos