In live video streaming, the size of manifest grows linearly as the overall time duration of manifest increases in many scenarios. Such a behavior exists across streaming technologies, e.g. Dynamic Adaptive Streaming through HTTP (DASH), HTTP Live Streaming (HLS) and Microsoft Smooth Streaming (MSS). It introduces significant overhead for manifest generation on cloud services, download latency and network traffic, manifest parsing and manifest storage on both cloud services and more important on devices with limited memory and CPU power, especially in the scenario of live video streaming where manifests have to be refreshed every several seconds. In this paper, we analyze what the problem is, where it comes from according to the basics in audio and video compression properties, and why it exists across the industry of live video streaming. We also look into a couple of case studies from some content providers and popular streaming services. Then we propose the novel pattern template, which can optimize the size of manifest, completely eliminate the factor of linear growth in time about the size of manifest, and make it stay constant in the scenario of continuous streaming without discontinuities. On service side, it saves computation for manifest generation, storage and parsing and optimizes the network traffic. On device side, it optimizes the download and manifest refresh latency, manifest parsing and storage in memory. The pattern template is applicable to any adaptive bit rate video streaming techniques and protocols at network application layer, such as HLS and Smooth Streaming, though this paper focuses on DASH streaming technology. Recently pattern template manifest has been adopted in DASH standard 6th edition in 2024.
Research areas
