In previous blog posts we addressed how optimizing the delivery of video over mobile broadband is a key to providing a good user experience and ensuring efficient use of network resources. The tendency of viewers to abandon videos mid-stream is one aspect of user behavior that mobile operators can exploit to increase network efficiencies as demand for video increases.
We are all guilty of attention-deficit surfing behavior: randomly bouncing among videos on popular social networking, news, and entertainment websites, rarely patient enough to view entire clips. Since the video server and client device tend to deliver and buffer the video well ahead of the viewing rate (also called the encoding rate) of the video itself, video abandonment can result in a waste of precious bandwidth and significant amounts of data dropped along the way.
Our traffic studies of live HSPA+ networks in 2011 confirm the dramatic effects of this user behavior. The following charts show the behavior of users streaming video sessions from a single base station. The analysis in this case is based on Flash video, the dominant video encoding format today.
The distribution chart above reveals the following:
- Only ~35% of videos are watched to completion (far right vertical line)
- 50% of videos are viewed for less than 40% of their total length
- 20% of videos are viewed for less than 10% of their total length
These trends point to wasted bandwidth when the encoding rates are compared to the transport rates for these videos in the chart below:
- For this sample of videos, the range of encoding rates is rather small (~500 Kbps to 750 Kbps), while the transport rate varies greatly (~500 Kbps to over 3 Mbps).
- Only a small percentage of video sessions match encoding rate to transport rate (<20%)
- The median difference between the encoding and transport rates is ~500 Kbps, which implies that every abandoned second of a video represents 500 Kbits or 62.5 Kbytes of wasted capacity.
Further analysis shows that an average video session wastes almost 1 MB because of poor matching between the encoding rate and the transport rate and early abandonment of the session. This significant inefficiency leads to higher levels of congestion and a degraded user experience.
The industry and operators have recognized the need to better match the transport and encoding rates and "over-buffering" of user videos - especially in times of high network use and congestion. Two approaches are currently being implemented:
- Video Pacing - This involves the use of various techniques to better match transport rates to the encoding rates of streaming videos. By more closely managing the user buffer depth, less instantaneous transport bandwidth is required and fewer bytes are wasted if a user abandons the video. Video pacing techniques include buffering and rate-shaping in the network, split-session video servers, and TCP session optimization to rate shape the streams. These techniques require specific knowledge of the video meta-data, client and server capabilities, and network conditions in order to prevent negative user impact (for instance, choppy or stalled video resulting from buffers not deep enough to accommodate network variability).
- Adaptive Streaming - This is separate and distinct from video pacing in that it attempts to provide the best user experience (highest video quality with lowest stalling/stopping of videos) by adaptively changing the encoding rate based on the available transport rate to the client. This approach is being actively pursued by the dominant video player vendors (Microsoft Silverlight, Adobe Flash, and Apple QuickTime) and standards bodies (HTML5). Major Internet video sources such as Netflix.com and Hulu.com already support adaptive streaming. In its most basic form, the adaptive streaming video client monitors its buffers to determine if the available network bandwidth (transport rate) is sufficient to support the video stream encoding rate. If the transport rate is low (i.e., the buffer is draining too quickly), the client requests a lower quality stream from the server to better match the rates. Correspondingly, if the transport rate is high (buffer filling too quickly), the client requests a higher quality stream.
In some respects, these two schemes are in conflict. Video pacing takes the network perspective and attempts to optimize the network resources needed to deliver a video stream whereas adaptive streaming takes the user perspective and attempts to deliver the highest bit-rate quality for a given network condition. Nevertheless, the two techniques can be used together to provide the best overall user experience for a collection of users (not just a single user) in the face of constrained network bandwidth. Video pacing and adaptive streaming represent just a couple of the many tools and techniques operators can use to better manage the unpredictable data behavior in their networks. Without content and flow optimization at critical points in the network, congestion and its resulting user impact will occur more frequently and lead to expensive over-building of the network and costly subscriber churn.