May 20, 2022 by Romain Bouqueau

GPAC is the fastest CBCS encryptor

Introduction

Our industry faces several challenges: extending from video to attention time (video games, ...), following users' ever-changing habits that escapes the traditional business (TikTok, Social Media ...), being greener. This implies more flexible and efficient workflows.

In an effort to use less resources and accelerate the processing, Comcast and Intel contacted Motion Spell to accelerate GPAC for CBCS encryption processing. This was based on an on-going effort from Intel to promote their new AVX SIMD instructions.

Mile High Video

Mile High Video is a quite new conference (see also our contribution at the last edition). It's been promoted as an ACM conference this year, a label that confirms its premium quality.

Technically we profiled GPAC. We know that fully loaded computers are more energy-efficient. This is equally true for the SIMD vectors. All the challenge was to improve the scheduling for both improving the load and the processing time.

Results and details

We have considered some typical Comcast content for our benchmarks. For the publication of the poster, the following content were used:

  • Big Buck Bunny: Full HD@30fps (3Mbps), Full HD@60fps (4Mbps), 4K@30fps (7.5Mbps), 4K@60fps (8Mbps)
  • Chimera: 4K@24fps (54.4Mbps)

The experiments ran on a Intel Xeon Gold 6348 CPU on on Ubuntu 20.04.03 LTS.

The time spent within encryption went down from 55% to 90% compared to the best implementation (based on OpenSSL). The overall runtime went down from 25% to 65%:

Performance gains in GPAC between OpenSSL and the Intel MBSec-IP library

Download the full article from the ACM Website: "Multi-buffer AVX-512 accelerated parallelization of ‘cbcs’ common encryption mode".

An unexpected source of latency (part 2)

previous post

<

Thriving open-source ecosystems: the DVB OSMART workshop

next post

>
  • - Comments are closed -