New Data Released On The Performance Of Adaptive Streaming Over HTTP
Adaptive streaming over HTTP is gradually being adopted by content owners as it offers significant advantages in terms of both user quality and resource utilization for content and network service providers. To understand whether today's existing commercial players perform well, especially under dynamic network conditions, Cisco and The Georgia Institute of Technology just released a technical white paper on the subject. Their experiments covered three important operating conditions:
- First, how does an adaptive video player react to either persistent or short-term changes in the underlying network available bandwidth? Can the player quickly converge to the maximum sustainable bitrate?
- Second, what happens when two adaptive video players compete for available bandwidth in the bottleneck link? Can they share the resources in a stable and fair manner?
- And third, how does adaptive streaming perform with live content? Is the player able to sustain a short playback delay?
The paper identifies major differences between Microsoft's Smooth Streaming, the player used by Netflix, and one open source player (Adobe OSMF). Their findings report that the Smooth Streaming player is quite effective under unrestricted available bandwidth as well as under persistent available bandwidth variations. It quickly converges to the highest sustainable bitrate, while it accumulates at the same time a large playback buffer requesting new fragments (sequentially) at the highest possible bitrate.
On the negative side, the paper says that the Smooth Streaming player reacts to short-term available bandwidth spikes too late and for too long, causing either sudden drops in the playback buffer or unnecessary bitrate reductions. Further, the experiments with two competing Smooth Streaming players indicate that the rate-adaptation logic is not able to avoid oscillations, and it does not aim to reduce unfairness in bandwidth sharing.
The Netflix player is similar to Smooth Streaming as they both use Silverlight for the media representation. However, the paper reports that the former showed some important differences in its rate-adaptation behavior, becoming more aggressive than the latter and aiming to provide the highest possible video quality, even at the expense of additional bitrate changes. Specifically, the Netflix player accumulates a very large buffer (up to few minutes), it downloads large chunks of audio in advance
of the video stream, and it occasionally switches to higher bitrates than the available bandwidth as long as the playback buffer is almost full. It shares, however, the previous shortcomings of Smooth Streaming.
The paper says that the OSMF player often fails to converge to an appropriate bitrate even after the available bandwidth has stabilized. This player has been made available so that developers will customize the code including the rate-adaptation algorithm for HTTP Dynamic Streaming for their use case.
There is a lot of technical details in the paper and it makes for a good read if you are interested in the adaptive streaming topic. If you have a question on the paper, leave it in the comments section and I'll ask one of the authors to answer them.