New Currents in Streaming

Painful as it has been for investors, the dot com dive in financialmarkets has been a useful reminder that doing it over the Internet– whatever “it” is – doesn’t allow you toignore the fundamentals of your chosen field. If commerce is your game,your days are numbered unless you manage to take in more than youspend. If your field is communication, there’s anotherfundamental you’d best not ignore: Your medium won’t holdyour audience unless it provides acceptable “quality ofexperience.”

Along with compelling content, a medium’s convenience, cost,and audio-visual resolution all shape the end-users’ perceptionsof their experience. Today’s popular media exhibit each of thesequalities in varying proportions. Broadcast television, for instance,is inexpensive, and may (depending on your local topography) providepassable audio-visual quality. But it’s not really convenientunless the content you want to watch happens to be on precisely whenyou want it

Video streamed over Internet Protocol networks, on the other hand,holds out the tantalizing prospect of true video on demand, with anyprogram just a click or a remote-key away at any hour of the day. Thatcovers convenience, but it doesn’t address the issue ofaudio-visual resolution. There’s a minimum threshold below whichit is nearly impossible for viewers to suspend their consciousness ofthe medium and concentrate instead on the content. Video streamed overbroadband connections (DSL, cable modem, Ethernet) may meet thiscritical test, but video over dialups does not.

For the foreseeable future, the vast majority of home Web users willaccess the Internet via dialup. That doesn’t mean that streamingvideo is dead on arrival. But video professionals intrigued bystreaming’s potential should probably take a realistic look atwhere the opportunities really lie in today’s market, and alsokeep an eye on how developments in technology may influence the marketsof tomorrow.

Today’s market

In terms of today’s streaming video market, at least twodevelopments bear watching. The first is that more and morenon-consumer Internet users – corporations and institutions– have the bandwidth for streaming, not only within theiroperations (intranet) but also to and from the Internet (broadbandconnections). This makes the corporate market the prime growth arearight now for services related to video streaming.

“For now, the bandwidth is in the corporation,” saysScott Gordon, vice president of marketing at SeeItFirst in Fremont,California. “And we have seen rapid adoption of streaming videoin the corporate space. The investment is based on theapplication’s cost-benefit. Applications include webcasting largecorporate events, training internal personnel, informing shareholders,training sales partners, extending customer service and support,launching products, and enhancing human resource initiatives.It’s basically about efficiently disseminating important andtimely information in a captivating and interactive medium.”

Another important trend is that broadband to the home, while still asmall piece of the pie, is growing rapidly. The audience size requiredto make entertainment-oriented video streaming a viable proposition maybe a ways off, but it is coming.

“Don’t let the current number of broadband-enabledconsumers hide the underlying trend,” says Greg Lowitz, generalmanager of the Webcasting Solutions Group at Pinnacle Systems inMountain View, Calif. “The demand is there, but the companiesproviding the service are working through the issues that face anystart-up business or industry as it scales to meet that demand. Thefact is, consumers want broadband faster than telcos and cablecompanies have been able to roll trucks to the home.”

Scaling redefined

Even as deployment of broadband expands, millions of users willcontinue to access the Internet via dialup, and adoption oflow-bandwidth wireless devices will continue to grow. That means thedemand for ever more efficient ways to utilize existing bandwidth willcontinue. Compression is key, of course, and we can expect furtheradvancements in the existing codecs used in RealVideo, QuickTime, andWindows Media. But it seems unlikely that these codecs alone can evermake low-bandwidth video truly palatable. Instead, performance willadvance through a combination of technologies, some existing and somenew.

One key development will be a redefinition of the concept ofscalability, enabled by the MPEG-4 standard. A comprehensivedescription of the standard, which is designed to facilitate thestreaming of a wide variety of media content at bit-rates ranging fromnext to nothing up to 10Mb per second, is beyond the scope of thisarticle. But with MPEG-4, it becomes possible to move beyond the simple“bandwidth negotiation” of today’s media players, inwhich the server determines the bandwidth of the connection anddelivers a stream that has been pre-made to match one of a few commonbit-rates.

MPEG-4’s scalability operates simultaneously at severaldifferent levels, and on several different parameters. To understandit, you first have to know that MPEG-4 – like QuickTime, on whichthe file structure is based – was conceived to allow theintegration of multiple synchronized media streams into a unified“container” format. The streams may be visual (video,animation, etc.) or aural (digital audio, synthesized music, or speech,etc.). These individual streams, along with associated metadata abouttheir time-base, duration, etc., become “media objects.”The overall MPEG-4 file of which these objects are a part is not itselfactually streamed.

Another key MPEG-4 concept is that the exact presentationtransmitted to and experienced by a given end-user is bothcontext-dependent and interactive. Media objects maintain theirindividuality within the MPEG-4 container. The way the objects arecombined at playback is defined by “scene descriptions”that use a format based in part on Virtual Reality Modeling Language(VRML). These descriptions can be conditional, meaning that thecomposition of a scene may depend on the playback context, includingboth the connection over which the scene is transmitted and the deviceon which it is played.

At the scene level, MPEG-4’s scalability means that the numberof objects used to compose a scene may vary depending on bandwidth.Rather than trying to cram everything through the pipe at once, thecomponent parts of a scene may be intelligently prioritized based onhow essential they are for conveying meaning.

A news show scene, for instance, may be composed of an announcershot against blue-screen, the announcer’s voice, a motionbackground (the busy newsroom), an inset video window for news footage,a character-generated headline strip across the bottom, and abackground music bed. A viewer watching a television hooked to adigital set-top box might see and hear all of these objects. Over a100Kbps connection to a PC, perhaps just the news footage would betransmitted, along with the announcer’s voice. The wireless phoneuser, meanwhile, would get only the announcer’s voice.

In addition to this scene-based scaling, MPEG-4 provides for each ofthe individual media objects to be scaled based on a number ofparameters. In spatial scalability, the screen resolutions used todisplay textures, images, and video objects adapt to availablebandwidth. Temporal scalability changes the displayed time-base (framerate).

Perhaps the most sophisticated is quality scalability. The encoderdivides the incoming signal into multiple layers: a base-case layerthat must be transmitted, and enhancement layers that may betransmitted as bandwidth allows. Each successive enhancement layerfills in detail and “richness” that progressively improvesthe experience as bandwidth increases.

Finally, to ensure that MPEG-4 can be affordably implemented acrossthe spectrum of consumer and professional devices, the formatincorporates complexity scalability. Essentially this means that whilethe best playback is obtained from a sophisticated decoder design, theformat is also designed to allow use of simple decoders, while ensuringthat the results are still comprehensible.

The distributed network

It’s early yet in the development of tools allowing contentcreators to take full advantage of MPEG-4’s innovations. Onceimplemented however, MPEG-4 will advance streaming by allowing a muchmore refined approach to making the best use of each individualconnection’s bandwidth. In the meantime, a couple of otherapproaches to better bandwidth utilization are already available. Oneconcept that’s received a lot of attention is distributed networkarchitecture. According to Akamai, a leading proponent of thisapproach, the infrastructure of the Internet involves several inherentbottlenecks.

Describing the Internet as a “network of networks,” thecompany says the speed and reliability of data transfer are affectednot only by the capacity of the system’s backbone networks, butalso by the limited number of “peering points” at whichthese networks hand off data. The multiple routers and backbonesthrough which data normally travels can cause delays and packetloss.

Akamai also describes a “first mile” bottleneck, whichreflects the fact that most websites disseminate data to all userseverywhere from a central location. That limits the speed with whichmultiple users can simultaneously access a site’s data to thebandwidth of the site’s single connection to the Internet.

Akamai’s solution to these problems is to replace centralizedcontent serving with what the company calls “edgedelivery,” in which a company’s content is“cached” on server sites spread out across multiplebackbones. Branded as FreeFlow, the idea is to shorten the path takenby the data on its way to the client, thereby skirting the first mile,peering, and backbone bottlenecks. To make it work, Akamai currentlyoperates more than 6,000 servers on more than 335 networks in 54countries. The company uses realtime monitoring of Internet trafficflow to map the fastest route to a given end-user, and to re-route inresponse to congestion and network outages.

A burst of video

An alternative approach to efficient bandwidth utilization ischampioned by Burst.com. The company offers its Burstware products as asolution for what it says are shortcomings of the two main methods usedfor streaming on the Web today.

Describing the realtime streaming used by proprietary videostreaming servers, such as those offered by Real Networks andMicrosoft, Burst says audio and video data is transmitted at a constantbit-rate, recorded into a buffer on the client machine, and played atthe same rate at which it was transmitted. To allow the buffer to fillwithout a long delay before playback starts, the buffer is kept verysmall. When fluctuations in the network environment prevent a constantflow of data, the buffer is quickly depleted, and the data delay orloss shows up on screen. To minimize the problem, the bit-rate has tobe limited to the minimum bandwidth one can expect to achieve over thetype of connection being supported.

In HTTP streaming, meanwhile, the bit-rate of the streamingisn’t limited by the bit-rate of playback; the buffer fills asquickly as the connection will allow. However, there’s noprioritization of the way in which multiple requests for streamed dataare filled. An HTTP server attempts to deliver each client as much dataas possible as quickly as possible until delivery is complete. Thatmeans the client with the biggest pipe can hog a big chunk of firstmile bandwidth, without regard for the effect on any other clients.

According to Burst, content providers using Burstware avoid theseproblems by “bursting” data from their servers to theirclients. A “Conductor” component monitors server activityand distributes bandwidth demand evenly, avoiding overload. Theservers, meanwhile, continuously monitor the buffer levels andavailable bandwidth of each client. Whenever bandwidth is available,Burstware replenishes the client-side buffers with managed“bursts” of content, prioritizing those clients whosebuffers are nearest to depletion. That allows the clients to play forlonger out of their buffers when demand for video bandwidth is high ordata flow is interrupted. It also means that the video can safely beencoded at the average bit-rate supported by the connection type,rather than just the minimum bit-rate, which in turn allows higherimage quality.

The catch is that a media player must be “Burst-enabled”to work with a Burstware server. So far, Burst has made availableBurst-enabled versions of Windows Media Player and the QuickTime forWindows Player. The system works with any media type supported by theseplayers, including MPEG-1, .avi, .mov, H.263, and Sorenson Video.Developments like Burstware, Akamai, and MPEG-4 highlight one of thegreat strengths of the Internet as a delivery medium: the flexibilityof a system that is implemented primarily in software. True, thehardware infrastructure has to be in place to make media deliverypossible, but it doesn’t have to be ripped out and replaced toallow improvements that can boost the quality of the end-user’sexperience. That bodes well for everyone who’s ready to investtheir content creation skills in the evolving world of Net media.

Phil DeLancie is a freelance writer based in Berkeley,CA.