Video presents a unique conundrum for mobile service providers. On the one hand, the ability to deliver video efficiently to mobile devices presents a significant opportunity for new services, revenue generation, and competitive differentiation. Services such as interactive video, anytime and anywhere delivery of premium video content, and real- time surveillance are only a few of the possible value- added services that rely on efficient delivery of video to mobile devices.
On the other hand, the sheer volume and growth of video traffic is straining mobile operators’ networks to their breaking point — making it difficult to deliver video with the quality that users expect. According to Yankee Group, video is projected to exceed 65 percent of all mobile data traffic by 2013. In December 2010, YouTube announced that each day 200 million videos were being played on mobile devices globally – up 300 percent from a year before.
As mobile access becomes more ubiquitous and devices more capable, subscribers will be less likely to tolerate degraded quality—rather they will expect an experience similar to that which we now enjoy on wired networks (much as users did in the early days of mobile voice; the expectation was based upon many years of reliable landline services). This means that to capitalise on the growing demand for mobile video, operators must first optimise their networks to deliver video efficiently, at scale, and with exceptionally high quality.
There are several options available to operators to address the growth in data traffic, including: Wi-Fi offload, introduction of HSPA+ or LTE, more base stations, more spectrum, and video optimisation technologies. In order to meet the demand and expectations, a combination of all of these will be required. This article focuses on technologies which can optimise video streams before they hit the most expensive part of the network: the radio access network (RAN).
A number of tools are available for mobile video optimisation, which will be described in the following sections. Again, the best results will be achieved by applying a coordinated combination of all these mechanisms.
Video streaming technologies
Any discussion of video optimisation must start with a discussion of streaming technologies. The two most popular video streaming technologies are progressive download (PD) and adaptive streaming (AS). Both are in common use today and require very different optimisation approaches.
Progressive download is used by most web sites on the Internet (including YouTube). It downloads a video file at the fastest rate that can be supported by the network. Once the mobile device has enough of the file in its buffer, it begins to play the video. There is no correlation between the download speed and the playback speed. PD counts on a deep buffer in the client device to handle any changes in network performance. Such changes are very common in a mobile environment where radio performance will typically drop way off as the user gets near the edge of a cell and pick up as the user gets closer to the cell center (i.e. the base station).
A different approach, adaptive streaming (AS), is starting to replace progressive download at many Internet sites. With AS, “chunks” of content, usually representing a few seconds worth of video and audio, are streamed one by one to the mobile device. When the mobile device is ready, it will request the next chunk. If network conditions change as the user moves around, the client can request that the next chunk be downloaded at a lower or higher bit-rate. Adaptive streaming videos are stored on the origin server using a number of alternate encodings, at a variety of different bit-rates to handle a variety of different network conditions. Streaming at a different bit-rate provides very much the same function as pre-filling a large playback buffer in PD, but in a much more bandwidth efficient way.
The primary method for optimising progressive download traffic is to compress it. There are several ways to compress PD traffic. The most popular is content-aware compression, where the algorithm takes advantage of the way the human brain processes video and removes data that would not be noticed. This technique can consistently compress traffic by about 30 percent with almost no impact on the user experience. Higher compression ratios are also possible, but that becomes more noticeable. Even then, compression offers a superior user experience compared to videos freezing and stuttering during times of congestion.
Another technique that can be used to optimise PD content is just-in-time delivery (JIT). Unlike compression, this is a lossless technique. JIT can also be thought of as buffer management. Instead of allowing the mobile device to download a large PD file and then start the playback function, content is metered out slowly so the mobile device’s buffer only has just enough data to enable smooth playback. The reason this approach provides savings is that most Internet videos are not viewed to completion. If the user abandons a video after a few 10s of seconds, all of the data left in the device’s playback buffer is wasted. With JIT delivery, the size of the playback buffer is carefully managed to minimize this waste. Of course the downside to JIT is that, in case of sudden degradation of bandwidth, the buffer will empty and playback quality will suffer....
As part of the setup procedure for Adaptive Streaming (AS) videos, the origin server will send the mobile device a manifest that lists all the bit-rates that can be supported when streaming that video. The mobile device selects one of the offered bit-rates and the process begins. The tendency is for the mobile device to select the fastest bit-rate that it can handle, but this may not be in the best interest of the mobile operator.
One technique to control this process is to modify the manifest so that the mobile device never sees the higher bit-rate options. If the choices are 100 kbps, 200 kbps, 500 kbps, and 1 Mbps it might be in the operator’s best interest to remove the two higher speeds from the manifest and let the device pick from either 100 kbps or 200 kbps. Smartphones do not need more than 100kbps, but if you are viewing a video on a high-end laptop with wireless dongle, then the video quality will be lower than optimal.
Once the flow is under way, the manifest can’t help with further modifications of AS content. In that scenario traffic shaping comes into play. In traffic shaping the network will slow the flow down to the rate that it can handle and this will “encourage” the mobile device to select a lower bit-rate from the origin server.
While video quality may be somewhat impacted (especially for PD), this is better overall than allowing the congestion to continue.
Certain videos can become very popular and sweep across the Internet. In this scenario, it makes great sense to cache these videos locally in an optimised format. Then all future users who wish to access the video will get a much quicker download than if it were necessary to fetch it from the origin server. In the case of PD streaming, these videos would be cached in an already- compressed format. In the case of adaptive streaming,
multiple low-bit-rate versions would be brought down from the origin server and cached. DRM-encoded content (Digital Rights Management) should not be cached and would always need to be fetched from the origin server. When using a cache, it is always necessary to check with the origin server each time there is a cache hit to make sure the content is still valid (hasn’t been pulled for copyright reasons) and so the origin server can keep count on how often a video is accessed.
Video optimisation is most appropriate when used for congestion management. The goal for service providers is to maximise the return on their RAN investment, while at the same time keeping users happy. A compelling user experience is only possible if congestion can be avoided – either by video optimisation or by increasing RAN capacity.
With a properly-designed video optimisation toolkit, it is possible to perform optimisation only in those parts of the network where it’s really needed. A single, centralised optimisation function can thus avoid or mitigate congestion across an extensive RAN footprint.
While some forms of optimisation may result in reduced video resolution, in most cases users won’t notice – certainly compared with not doing any optimisation at all!
The end result for subscribers will be a high quality user experience, while service providers are able to defer investments in additional RAN infrastructure – potentially resulting in a return on investment (ROI) of nine months or fewer.
With mobile video optimisation, the end result for subscribers will be a high quality user experience, while service providers are able to defer investments in additional RAN infrastructure – potentially resulting in a return on investment of nine months or fewer.
About the authors
Gijs van Kersen is Marketing Manager for Mobile Solutions, Juniper Networks, in the EMEA region. Contact: firstname.lastname@example.org
Steve Hratko is Marketing Manager for Mobility Products, Juniper Networks. Contact: email@example.com