|
Feature
April 2000
Hot & streamin'
The online masses want moving pictures. Now, if you please. We dive into streaming video.
Brian Dipert, Contributing Editor
Back in the early days of the Web (like, uh, five years ago), early-adopters surfed just because it was the latest thing. But the novelty of humble HTML quickly wore thin. And now the Internet audience has grown to include the average John and Jane Doe, folks who aren't enamored with high-tech for high-tech's sake. They want an experience that derives from, and expands on, the television content that they already know and love. They've heard the promises of thousands of video broadcasts streaming over the Internet. They're going to expect nothing less.
There's just one little problem with this build-on-the-TV strategy. Very few of us enjoy a personal T1 line like Bill Gates. Most net users still log on from home with slow analog modems, which choke on bloated multimedia data rates. Yet John and Jane Doe aren't going to be happy with a postage-stamp-sized video presentation that reminds them of Max Headroom.
Since George Gilder's vision of fiber optic cable to the masses hasn't yet become a reality, multimedia technology providers have been forced to develop some pretty clever workarounds. Steve Jacobson, program manager at RealNetworks, summarizes the basic goal of streaming multimedia pretty simply, "The two most important things to a content developer are that the observer be able to a) see the content, and b) take an action based on the content." The devil's in the details, though, and economic realities might ultimately point toward approaches that the technical mind would rate as second best.
A few words on terminology before we continue. Traditional multimedia delivery follows a download-and-play paradigm. You request the content from a server. The server sends it (temporarily or permanently) to your PC, the client. Then playback begins. More sophisticated client software can, by sensing download speed and comparing it to total file size, begin playback before it captures the entire file, giving the illusion of streaming delivery. Regardless, you're still cluttering up your hard drive with files that, in many cases, you won't even care about as soon as you've seen them once.
|
John and Jane Doe aren't going to be happy with a postage-stamp-sized video presentation that reminds them of Max Headroom. |
|
This download-and-play model has several limitations. First off, it assumes that the client has sufficient storage space to hold the entire multimedia file. This supposition may be reasonable for today's PCs and their multi-gigabyte hard disks. But it breaks down the moment you go beyond the PC, such as to a digital set-top box or an Internet appliance. Secondly, content creators don't want consumers to be able to snag a permanent copy of the file. Even if the file is encrypted or watermarked with media-security protocols, time and ingenuity can surmount those precautions. And any halfway-competent computer user knows how to move information from the browser cache to a more permanent location.
More significant, download-and-play also precludes live broadcasting, specifically multicasting. This one-to-many topology is efficient both in terms of network bandwidth and server resources. The server doesn't have to chronologically service every incoming request for the content. It simply broadcasts the stream to the entire viewing audience.
The bigs
Any streaming multimedia discussion would be remiss if it didn't begin with today's big three providers: Apple (QuickTime), Microsoft (Windows Media Technologies), and RealNetworks (RealSystem). Although this battle of the titans resembles the browser war of the last few years, the outcome may be far different, even without Department of Justice intervention. All three companies supply tools for rudimentary content editing, encoding and transcoding, server-side delivery, and client-side playback. But beyond these high-level similarities, key differences in strategy and implementation quickly emerge.
Apple's situation is ironic, because this multimedia pioneer was the last of the three to add streaming capability, with QuickTime 4.0 in June of 1999. Not until this spring's version 4.1 did QuickTime incorporate advertising insertion and support for SMIL (synchronized multimedia integrated language)—a standard that allows the synchronization of diverse media types in a single presentation. But in a twist seemingly inspired by Linux, closed-architecture Apple last summer embraced the open-source model, giving QuickTime the means to extend its server-side reach beyond the Macintosh and Apple OS markets.
Don't misunderstand: Apple's first priority is still to sell as many copies of its OS X software and Macintosh G4 hardware as possible. The company bundles its QuickTime Streaming Server with the server OS, and is quick to remind potential customers that unlike RealNetworks, QuickTime doesn't require payment of a per-stream "tax." But developers have also ported Darwin (the code name for the open-source version of Streaming Server) to FreeBSD, Linux, and Solaris, with other ready-to-use binaries on the way.
|
As much as the streaming media providers like to talk about standards, their products are based almost completely on proprietary file formats and network protocols. |
|
Like Apple's open-source partners, RealNetworks supports multiple OSs with its server software. Depending on which RealServer version you're interested in, the list includes Windows NT, 9x, and 2000, the Macintosh OS, FreeBSD, and a number of Unix variants including Linux. Unlike Apple, however, RealNetworks isn't giving the software away, except in versions that put limits on functions, support, and user configurations. The company's free Basic Server supports up to 25 concurrent client sessions, the $1995 Plus version serves up to 60 users, and Professional comes in versions for 100, 200, 400, 1000, or 2000 users. Professional also supports network and third-party security extensions, as well as advertisement insertion.
RealNetworks follows a similar multi-tier strategy with its content-creation software. The free RealProducer Basic supports a restricted set of bit rates and codecs, while the $149 Plus version, among other capabilities, generates files that are backward-compatible with prior-generation client-side players (more on this later). The $399 Pro package builds on Plus, adding a number of pre-formatted HTML and SMIL templates, plus batch processing capability. RealNetworks has little interest in developing its own media-security protocol, but is content to partner with third parties such as a2b music, IBM, Intel, and Liquid Audio.
Take RealNetworks' strategy, imagine the exact opposite, and you've pretty much captured Microsoft's approach to streaming multimedia. Microsoft lets you freely download its content-creation, server, and media-security software, and even bundles Media Services with Windows 2000. But your no-cost license to use these tools is only valid if you host your content on Windows NT or 2000 servers. Microsoft's ultimate goal is to sell server licenses. In the words of product manager Sean Alexander, "the mission of the digital media group is to support the operating system."
Attempts to equate past browser battles between Netscape and Microsoft to today's multimedia marketshare tug-of-war are valid to a point. Yet there are important differences. When Microsoft and Netscape were at war, Linux wasn't yet a major competitor to Microsoft in the server space (I don't count Solaris because its free license is only for non-commercial use). Now, Microsoft's competitors can argue that their tools are cheaper to implement, because they run on low-cost or no-cost Linux, whereas Windows Media is more expensive because of the price of the OS, particularly Windows 2000's expensive Advanced Server version. Also, as much as the streaming media providers like to talk about standards, their products are, unlike a web browser's HTML, FTP, and HTTP, based almost completely on proprietary file formats and network protocols. Once you've committed to one vendor's technology, you face high barriers moving your content elsewhere.
Microsoft bundles a full-featured media player with every OS it ships, yet in Windows 98 it also provided the option of installing a now-obsolete version of RealNetworks player. The relationship between the two companies dissolved, however, so don't count on seeing this action repeated in future Microsoft OSs. Also, neither Apple nor RealNetworks wants to repeat Netscape's mistake of attempting to make people pay for the client software. Although both companies offer a sub-$30 upgrade to their freeware players, the added features aren't compelling or necessary for most users, and neither company sees upgrades as a critical revenue source.
The clever client
Whether unicast or multicast, the streaming-media technique, in which the data is decoded, presented, and then immediately discarded, presents a larger technical challenge than the download-and-play approach. First off, the client doesn't normally have time to request a resend of a network packet that it doesn't receive. Instead, the client must gracefully degrade its presentation quality. The player must also buffer the incoming data stream to resolve both transmission delays and out-of-order packet reception. Finally, the data stream often consists of multiple media types that the player has to synchronize. These include still images, video, sound files that are based on different codecs and have different sampling rates, and text in HTML, XML, and other formats.
Each client must also communicate its available bandwidth (analog modem, DSL, T1, etc) to the server. Suppliers differ on their approach to this issue. Both the QuickTime Player and RealPlayer ask you to specify a default bandwidth when you install the software. This is fine, unless, like me, your notebook computer hooks up to a fast pipe in the office but uses a slow analog modem on the road. In Microsoft's approach, the client and server negotiate for the optimum bit rate when they connect. However, many sites that use Windows Media offer you multiple versions of a given media clip, based on your download speed. According to RealNetworks, this is necessary because each Windows Media stream supports only a limited number of simultaneous audio and video variants. RealNetworks adds that it's better to select the bit rate once, at installation time, because most computers access the Internet through one consistent means.
With the release of QuickTime 4.1, all of the big three now offer the ability to monitor the download connection and, if necessary, temporarily throttle down to a lower bit rate. This prevents what users perceive as the cardinal sins: jerky video and stuttering audio. All of the players support a variety of lossless and lossy codecs, and some will automatically download and install the appropriate codec when you try to play a specific media type.
None of the players supports the latest iterations of its competitors' proprietary codecs, though all will handle industry-standard alternatives like MPEG-2 video and MP3 audio. If the intended audience will access the streaming content through PCs, this codec obsolescence is a minor concern. Software upgrades are a way of life, after all. But what if the player and its codecs are burned into memory that can't be upgraded, such as ROM (read only memory). For example, Casio ran into this problem with its handheld PCs. The Windows Media Players embedded in the handhelds ended up unable to play certain audio files, and Casio had to install a patch. And Microsoft isn't alone in possessing this limitation.
No shortage of standards
Microsoft touts its approach as being "standards-based" (note the careful wording), yet admits that what it calls an MPEG-4 file format doesn't completely comply with standards. Microsoft based the format on a draft version of the MPEG-4 spec (MPEG-4 standardization efforts, in fact, continue to this day). Windows Media's video format, although based on DCTs (discrete cosine transforms) like MPEG, is proprietary, as is its audio compression scheme. Similarly, Apple touts its compatibility with RTP (real-time transport protocol) and RTSP (real-time streaming protocol), yet its Sorenson video codec uses wavelet techniques, and its audio codec is a proprietary creation of Qdesign. RealNetworks is in the exact same situation with its RealAudio and RealVideo schemes.
All of these approaches heavily compress the media before delivery and therefore require a reasonably powerful client to perform real-time decompression. Low-bit-rate versions supplement lossy compression with other common data reduction techniques such as reduced frame rate, limited color depth, and small frame size (otherwise known as low resolution).
One seemingly obvious answer to the obsolescence problem might be to base your streaming multimedia system only on rubber-stamped industry standards such as JPEG and MPEG. Yet this approach isn't foolproof either. For example, not all image viewing programs support the progressive-JPEG format, nor can all audio playback software handle variable-bit-rate MP3.
Another option is to base the client software on Java, as Geo Interactive has done with its Emblaze product. The key problem here is performance. Placing a Java virtual machine between the compressed media and codecs and the target OS and hardware further strains the system. Also, by restricting yourself to standard protocols, which lag behind the state-of-the-art by definition, you'll be turning your back on the quality, multiple-format synchronization, and interactivity that make streaming media compelling.
What if you've got a closed streaming-media system, such as a company Intranet or a cable television setup, where you're confident that all of your customers will use the same software or obtain their set-top boxes from a common source? In these cases, obsolescence isn't nearly the concern it might be operating across the broader Internet. If you're a hardware designer of a broadly-marketed set-top box, a multimedia-enhanced cellular phone, or some other full-featured Internet appliance, you might want to consider putting a hard disk in your client, or using reprogrammable flash memory to store the OS and applications, instead of write-once ROM.
Another criterion to keep in mind when evaluating client software is memory footprint. In a refreshing change of pace from Microsoft's typical reputation for code bloat, Media Player is generally regarded as lean and mean. In response, Real Networks highlighted the reduced memory footprint of its latest player, version 7. Important differences also exist with each vendor's current implementation of multicast, specifically the presence or absence of a unicast back-channel—a single pathway back to the server, allowing users to partake in activities such as buying stuff. For more technology details, visit the vendors' sites or attend their periodic seminars. Apple, for example, hosted its first QuickTime Live! Conference last fall, and RealNetworks' conference is coming up in May.
If you don't want to commit to a specific streaming platform, a number of third-party tools will let you encode content into multiple streaming-media formats. Popular options include Sonic Foundry's Stream Anywhere, Terran Interactive's Media Cleaner Pro and the various input and output plug-ins for Adobe Premiere.
|
Variations on a stream
Most of the popular attention paid to streaming multimedia focuses on the combination of audio and video (plus, perhaps, a bit of HTML or XML). But video may be overkill for your application. Or the bit rate required to produce a decent frame size and refresh rate might exceed the bandwidth capabilities of your target audience. In these cases, consider other means of delivering ear and/or eye candy.
If high fidelity audio-only delivery is sufficient for you, consider Liquid Audio. The company's streaming-media system uses high-quality audio codecs, includes robust media security and e-commerce capabilities and also lets you stream text and still images to the client. Plus, Liquid Audio's ultimate desire is to be a content provider, so the company recently concluded a partnership agreement with Microsoft. That bond will lead to versions of the Liquid Audio Player that support Windows Media codecs and security protocols.
In the 2D graphics space, Macromedia has made a name for itself with Flash. The Flash format uses vector graphics, which represent lines and curves as compact and scaleable mathematical equations rather than as boatloads of pixels. Flash version 4 supports streaming MP3 audio, and the server-side content is OS-independent. Flash files are, like HTML, dumb. That is, they require no server-side management. All the decoding and synchronization intelligence resides in the 150-kbyte Flash Player. The player is not only small, its source code is freely licensable and can easily be recompiled to various browsers and client-side operating systems, according to Macromedia.
Add interactivity to Flash, and, simplistically speaking, you have Shockwave. Again, the basic setup is server-independent (though the recently added multi-user capability complicates things a bit). However the Shockwave Player is between 1.5 and 2 Mbytes in size and hooks more deeply into the OS it's running on (meaning it's less portable to other OSs). While the streaming video competitors pummel away at each other, they're content (at least for now) to license and incorporate support for Flash and Shockwave within their players instead of developing an alternative technology.
Using the Shockwave development tool, Director, along with some third-party extensions, it's possible to create pseudo-3D environments, analogous to the game Doom (check out Real Pool at www.shockwave.com, for example). However, for realistic 3D, you need to look elsewhere.
Enter MetaCreations and Pulse Entertainment, companies whose delivery of compact, higher-level content descriptions for client-side rendering and display harken back to Microsoft's ill-fated ChromEffects project. MetaCreations uses a progressive-mesh approach to creating 3D images. The first few polygons streamed to the client create a low-quality 3D representation. Additional polygons refine and smooth the surface. The result is a browser-side presentation not unlike that of progressive JPEG. Pulse Entertainment approaches the challenge slightly differently. A one-time download stores each character's polygons, textures, skeletal structure, and deformation details to the client in a dedicated browser-independent cache. Virtual Jay Leno, for example, is only 80 Kbytes in size (it's also hilarious and disturbingly lifelike, I might add). Subsequent streaming includes audio, animation and synchronization data for the specific presentation. In other words, the character traits get reused.
To get a feel for the various technologies and their capabilities, compare the Dilbert content at www.dilbertondemand.com (Windows Media Technologies), www.shockwave.com (Macromedia) and www.dotcomix.com (Pulse Entertainment).
|
|
The long arm of the law
The Internet is often characterized as a "disruptive" technology, and nowhere is this trait more evident than in the area of streaming multimedia. Two examples should suffice to make my point.
iCraveTV used a loophole in Canadian law to capture US television broadcasts from across the border and rebroadcast them over the Internet, using a RealNetworks stream that included advertisements on all edges. Almost immediately, the company found itself staring at cease-and-desist lawsuits filed by broadcasters, movie studios, and other content providers, such as the National Football League. Shortly thereafter, a US court order compelled the company to at least temporarily cease its actions.
The plaintiffs claim two objections. First, they don't appreciate the degraded display quality caused by the streaming media conversion. Second, they say that iCraveTV didn't use sufficient security to ensure that only Canadian viewers could see iCraveTV's presentation (all a user had to do to gain access to the video streams was to enter a Canadian telephone area code). What's really bugging the plaintiffs is probably more like the following. They're unhappy that someone else is generating revenue from their material without giving them a cut. And the rebroadcasting places their core sales pitch to advertisers—regional targeting of content—in jeopardy. As this article went to press, the case was still tied up in the courts and iCraveTV's rebroadcast was still offline. For a humorous view of the whole proceedings, check out www.icraveicravetv.com.
Streambox.com got into hot water with RealNetworks when it figured out how to permanently capture and convert RealAudio and RealVideo streams to other file formats, such as WAV, MP3, WMA, and AVI. RealNetworks cried foul because the conversion took control of the media away from the content developer, and because the transcoding process degraded the quality. The fact that RealNetworks stood to lose per-stream revenue was probably also a factor. As of mid-February, the US District Court in Seattle had ruled that Streambox.com could continue to ship its audio-only Streambox Ripper program, but it forced the company to remove RealVideo capture capability from its beta Streambox VCR product.
Streambox.com vows to pursue its case in the courts, and in the interim to develop a production version of Streambox VCR that works with QuickTime and Windows Media streaming formats. Interestingly, it seems to be acceptable to capture audio only and to bypass media security; High Criteria's Total Recorder product does both, but has so far been ignored by the suppliers of streaming technology.
|
Author information
Contributing Editor Brian Dipert highly recommends that you check out "Cheesy Poofs" and the other South Park games at www.shockwave.com, "Virtual Bill" at www.dotcomics.com, and the "So Many Roads" Grateful Dead video at www.dead.net. But please, just say no to www.britneyspears.com.
|
