Guided just-in-time transcoding for cloud-based video platforms

Abstract

Consumption of streamed video on-demand (VoD) content is driving rapid growth of fixed and mobile network traffic. VoD services are commonly run on cloud-based video platforms, where all processing is software-based. The VoD content is typically delivered using adaptive bit rate (ABR) streaming techniques. As the amount of content in an asset library grows, the required storage space and the associated cost increase. This is emphasized if several ABR representations per content are stored. To save storage space, lower quality ABR representations may be eliminated and re-generated based on the corresponding high quality representation as the content is requested, which is referred to as just-in-time (JIT) transcoding. In this paper, we investigate a scheme that can significantly reduce the computational complexity associated with JIT transcoding, called guided JIT transcoding. We analyze its performance when multiple HEVC-coded ABR representations with different spatial resolutions are utilized, showing that for a configuration with seven video representations with resolutions ranging from 1080p to 360p, storage requirements can be reduced by about 24% while software-based transcoding from 1080p to 720p can be performed at 46 fps on average using a single execution thread.