With the industry's dynamic transformation to file vs. tape-based operations, solutions for effectively ingesting, managing and distributing digital assets are critical to a broadcaster's success. The variables in successfully delivering high-quality video over IP networks are numerous. Bandwidth optimization, compression choices, intrinsic latency, smoothing latency, scalability, integration with broadcast workflows and content delivery network (CDN) integration are just a few. And, while IP-delivered video may have a way to go to equal the experience available through traditional broadcast platforms, a quality Web-based viewing experience can significantly expand a broadcaster's reach, provide channels for potential revenue gain and set the table for success in the new media age.
This article discusses the challenges and solutions involved with the successful delivery of video over IP networks, with a focus on over-the top (OTT) video distribution.
Understand your network requirements
What kind of demands will OTT video — delivered to every conceivable playback device realized or imagined for the future — place on your network? That answer depends on how you respond to two key questions:
- First, what is your delivery strategy? That is, what content do you have to offer, and is it live, on-demand or a combination of both?
- Second, who is your audience? How many simultaneous viewers and what variety of playback devices do you expect? Which of those devices do you consider strategic to allow you to reach your goals? Are viewers predominately local, or do you have global aspirations?
The key factor that encompasses all of these variables, and that has the most impact on network capacity, can be reduced to one word: scalability — how many viewers, how many different streams and how much is to be stored for online access and playback.
Impact of live streaming
If your needs are simple, packaged systems are available that can encode, deliver and manage your entire Web video presence. (See Figure 1 on page 8.) These systems ingest real-time video and compress it to create and deliver the viewable IP streams in real time. In their simplest form, they create one stream (typically in Windows Media or Adobe Flash Live format) and deliver it to a small number of local users via your in-house media server. Many independent stations started their Internet presence in just this manner, and for a small number of stream types and viewers, this is still practical with little impact on the network and existing Internet connection.
However, with the explosion of interest in video streaming, increases in the number and types of devices used for playback, and a higher expectation to duplicate the video viewing quality of off-air broadcasting, the simple solution is no longer simple and no longer adequate in order to remain competitive.
Reaching out to a diverse online and mobile audience now requires multiple streams of differing characteristics. Where one or two streams sufficed in earlier times, the latest adaptive streaming technologies call for as many as five to seven or more streams from each video program source to serve even a small cross section of possible viewers. This essentially forces either a massive investment in in-house encoder and server resources, or forces an outsource-to-CDN strategy where you create separate uplink streams directed to a content delivery provider that, in turn, distributes your content to users or subscribers.
It remains possible to use a conventional Internet connection for one or two of those CDN uplink feeds, but at the risk of playback quality, reliability and consistency when using any but the most optimized Internet connection. As an alternative, consider using managed, unchallenged uplink feeds from a dedicated IP network service provider. They can guarantee reliable packet delivery to the CDN. Uplink quality is critical, any impairment of this feed affects all viewers everywhere, and recovery for even momentary network hits on the uplink are not as automatic as you might expect.
To enhance uplink performance, consider redundant streams on different networks. High-end media encoders can create multiple streams from the same video encoder, delivered via separate TCP/IP ports connected to different networks. Similarly, most CDNs accept multiple uplink streams and will automatically switch over when faults are detected. Of course, redundancy multiplies the impact on your network and will not be effective in preventing catastrophic outages if the redundant uplinks do not take diverse routes.
In practice, you will likely limit the Internet media output strategy to only one of several competing formats. The choice is often driven by the preferred development environment for your existing Web presence. If you already develop in Adobe Flash or use Adobe's Flex environment, Flash Live is a logical choice. If you develop in Silverlight, the Microsoft Smooth Streaming option is appropriate.
Both are examples of adaptive streaming, which offers a quality playback experience (including up to full 1080p HD) by creating multiple streams at different data rates simultaneously from the source content and delivering the one most appropriate for each individual viewer. That's good news to viewers because they experience minimal buffering time, fast startup time and an optimized viewing quality. But these new technologies multiply impact on your IP network and for most practical applications force a non-Internet uplink connection to a CDN. A typical adaptive streaming scenario includes as many as 11 simultaneous streams, although most installations use about five or less in actual practice.
Just how much bandwidth will live adaptive streaming demand of the network? Using Alex Zambelli's Microsoft Smooth Streaming bit rate calculator at alexzambelli.com/WMV/MBRCalc.html, to deliver five streams configured to reach a good cross section of users at a maximum of 1280 × 720 pixel playback frame size, create five streams of 400kb/s, 662kb/s, 1095kb/s, 1813kb/s and 3000kb/s. Frame size ranges down to 320 x 176 for some smaller devices. Bit rate supports 3Mb/s for well-connected users down to 400kb/s for mobile devices capable of receiving the same stream format.
But, most mobile devices in use today are not equipped to play the same stream formats used for normal Internet streaming for PCs, set-top boxes, game consoles and the new generation of network-enabled TVs. This is becoming less true as more mobile devices support Flash and Silverlight using H.264 compression, but for now there are several different compression technologies and distribution methods in common use.
iPhones are special cases. The encoder must create one or more H.264 streams (three are recommended), delivered as MPEG transport streams to a segmenter that in turn delivers chunked, time-sliced content to iPhone users. The segmenter may be included on the encoder or a separate service running on an in-house server, or on a server or server farm at your CDN.
At the moment, most video-enabled handheld mobile devices are limited by their network speed. A maximum rate of about 350kb/s-400kb/s is typical, configured with low bit rate mono audio in order to maximize the bandwidth available for video.
To have a broad reach to many viewers, create a large number of streams in real time, delivered via an unchallenged, well-managed, likely non-Internet-based network to a server farm (yours or those at a CDN), which in turn delivers an optimum experience to each viewer. If it seems a bit daunting to create and manage on your own encoder farm, there is another alternative: Backhaul a single full-bandwidth MPEG-2 feed to a company that specializes in transcoding and streaming in all desired formats. Yahoo! Broadcast, Level3 Communications and Real Networks are three examples of CDNs that offer such services to large network broadcasters.
Effect of on-demand video
On-demand video creates its own additional challenges for managing actual clips and segments and making them available to large audiences. Effective file-based delivery dictates that a similar number of different formats must be created and saved as files. The delivery impact on network bandwidth is similar at playback to the impact of live streaming, except that the number of simultaneous viewers is likely to be less.
Creating the different file formats is sometimes done by the live encoder as a byproduct of the live event. Many high-end encoders can be configured to save the live stream in the same media format for later playback. Other scenarios allow for the original source asset to be saved in uncompressed or lightly-compressed format and transcoded on the fly at playback as demand requires.
A future-proof comprehensive solution
The impact on the network to deliver on-demand video vs. a live stream is about the same, but hosting and managing a video library creates additional challenges to the IP infrastructure, particularly if you intend to offer long-term playback access to your ever-growing library of video clips. As the library grows, so does the number of simultaneous viewers.
Over time, you may find out that playback from the library exceeds the bandwidth consumed by the live stream operations. This is even truer if you create multiple different file formats for each video clip in order to handle several different playback experiences for different audiences. Furthermore, if you anticipate a need to repurpose the video assets, save the clip in an editable format such as MPEG-2. This creates a need for an even larger media storage array and a corresponding demand on the IP transport facilities.
One solution that can easily grow incrementally with your needs without periodic large investments is to outsource the entire video delivery operation, retaining the video capture and video asset management functions in-house. A comprehensive digital asset management (DAM) system can remotely manage the asset library as well as give you complete control over sourcing editing and access to content. Some DAM systems offer comprehensive features that help make live events available as quickly as possible after the event to on-demand viewers and completely automate ingest and delivery functions.
Whatever the size of your operation or the rate of transition to a new digital audience, there are many ways to select and configure a solution that will serve your needs well into the future.
Mark Hershey is vice president engineering for ViewCast.