Streaming video is now the principal concern of most network operators. I personally have a few YouTube channels that I follow, watch Netflix several times a week and keep up with much of what my friends post on FaceBook. The stats say I’m pretty typical.
For network operators this isn’t all good news; delivering and managing high-quality video streams is demanding for any kind of network, mobile especially. And whenever or wherever issues may arise, dissatisfied customers will always blame their access provider first.
One of the keys to handling customer dissatisfaction, and keeping ahead of your competition, is video analytics. With video being close to 50% of network traffic and growing, it’s critical to understand how it’s being consumed and whether your network is delivering the quality customers expect. Some OTT video providers (e.g., YouTube) actually rank operators by streaming quality, which is used by some consumers in choosing their operator.
There are several kinds of streaming technologies that have evolved over time, and some of the analytic approaches have not kept pace. Therefore, not all approaches to video streaming analytics are equal.
The easiest and oldest form of video streaming is HTTP progressive downloading. The device receives the video stream in chunks, which are buffered and cached locally for playback. Caching can be an issue for memory challenged devices, and it does not adapt the video resolution to the device’s processing, memory and screen size — called adaptive streaming. It also can’t be used for live streams. Most OTT streaming providers, such as YouTube, have phased out progressive downloads.
Real-time streaming protocol (RTSP) is a better protocol, but it relies on a special client and has a number of issues with getting through firewalls (NAT and port access), and runs on top of the UDP protocol, which makes it vulnerable to packet loss and jitter. Unsurprisingly, it only comprises 0.04% of mobile video.
The king of the hill, these days, is HTTPs adaptive streaming (HAS). As its name suggests, it adapts both to the bandwidth available and the client device (CPU and memory) for optimum playback. HAS is used by the big players including: Netflix, YouTube and Facebook, among others.
With HAS, the client browser or application keeps in synch with the streaming server using a manifest file, which is the first thing downloaded. It includes lots of details about the video stream, such as all the streaming bit-rates supported by the server, segments included in the stream and their addresses (URLs). The device starts out requesting the lowest bit-rate stream and then adapts upwards based on its own resources and the bandwidth quality. It adjusts by asking for different streams, depending on fluctuations in those parameters.
The problem with HAS is the little “s” for “secure” at the end of HTTP(s). With security enabled, not only is the video encrypted, so is the manifest file. As a result, legacy video analytics solutions that rely on parsing the manifest file and video payload are no longer feasible in providing analytics for the encrypted video. They rely on the manifest file to detect video flows, identify the video playout quality and determine the streaming KPIs.
Without understanding what the device is actually demanding and what metrics are or aren’t being met, the legacy video probes cannot alert the service provider to customer-facing issues that are a result of network performance. While optional, most OTT video streaming servers are now using the secure version of HTTP. Figure 1 shows the top video applications by volume and their breakdown by protocol; most of the applications are moving toward encryption by using HTTPS, HTTP2, and QUIC protocols.
There is, fortunately, a new generation of video analytics developed by Nokia that does not rely on the manifest file or parsing the video payload in HAS. The solution uses a novel approach that relies on complex machine-learning algorithms to identify video and audio chunks according to the traffic behavior. It can track the performance of a video session by correlating the encrypted chunks.
Using this information, the analytics are then able to detect the state and resolution of the video and apply KPIs, and this functionality is what is built into Nokia Cognitive Analytics for Mobile Networks. It ensures that subscriber performance issues are identified without waiting for phone calls to the help desk.
Have a look at our Cognitive Analytics for Mobile Networks infographic for more details.
Share your thoughts on this topic by replying below – or join the Twitter discussion with @nokianetworks using #analytics #bigdata #Machinelearning