Network Computingwww.networkcomputing.com

Compare Streaming Technologies
Windows Media Format	RealMedia Format
Click here to play	Click here to play
To fully appreciate the differences between these two media types, we invite you to view them using "full screen" mode. Note, you'll need the Windows Media Player and the RealNetworks RealPlayer to compare these streams.

We tested Apple's QuickTime (with the Darwin Streaming Server), Microsoft's Windows Media Technologies and RealNetworks' streaming system. We examined the characteristics of each streaming platform and assessed the quality of each implementation. We also tried out end-user content creation tools, explored options for media integration, and examined the ease of use and administration of encoders and servers. Setting up a simulated live broadcast from a static file and splitting that broadcast let us gauge the challenges of administering the server products and of scaling a real-time Webcast to serve many users across an enterprise.

Testing in the Stream

Our test environment included a network with a Dell Computer Corp. 400-MHz Intel machine with a ViewCast.com Osprey 100 video-capture card that was bootable as a Microsoft Windows NT 4.0 system for encoding. A separate Dell 900-MHz Intel box was set up as our primary media server. The encoder machine also was bootable as a Windows 2000 server with all three streaming media servers installed, letting us test splitting. Each player was installed on each system as well.

Although each streaming-media platform has unique strengths and weaknesses, they all work pretty much the same way. For on-demand streaming, you use encoder software to convert static media files into a highly compressed format suitable for streaming, then transfer the resulting files into a content directory on the streaming server. Users can then request and stream the resulting encoded content. For live Webcasts, you encode from video- and audio-capture devices on the fly and stream that content to the server, where it is transmitted in real time via unicast or multicast to users running the player software. Because live Webcasts are synchronous events -- unlike asynchronous, on-demand content, which can be viewed at your users' convenience -- they cause significant spikes in traffic, creating more stress for the server and the network.

The first thing we did was set up each server-encoder-player combination; we then created on-demand content using the Osprey 100 video-capture card, which converts a real-time analog video signal to an AVI (Audio Video Interleaved) video file. We captured at 240 by 180 pixels, converted to 192 by 144 pixels, fed those files into our encoders, copied to the appropriate server and streamed. Because the Osprey card cannot generate QuickTime or MPEG files, we used only AVI files for our tests.

We experimented with encoding multiple bit rates, and we set up live encoding for Real and Windows Media. We did not include the Apple product in this portion, because QuickTime Pro requires two pieces of third-party software for live encoding and multiple-bit-rate encoding, and neither was included in our test suite. (To do live streaming with QuickTime, you need Sorenson Media's Broadcaster software for Macintosh, and to encode at multiple bit rates, you need the codec for Sorenson Video Pro 2, now known as Video Developer 2.2.)

Integrating audio and video with other media -- for example, mixing video with images to create a slide show -- is an important aspect of many enterprise training and e-learning applications, and we looked at the tools for doing this in all three environments. We also examined the products' administrative and monitoring interfaces and features, and we created a simulated live broadcast from a static file and used it to test splitting.

Each vendor promises high quality at reasonable bandwidth, but RealNetworks, which offers the only server platform that isn't free, touted its ability to achieve higher quality at a very low bandwidth cost. The company asserts that its product is not as expensive as one might think because users can achieve good quality with less bandwidth, reducing the overall cost.

We hadn't planned on evaluating audio/video quality, a murky, subjective area. We're satisfied that quality video is achievable across the three platforms at reasonable bandwidths. However, RealNetworks' claims piqued our curiosity, so we captured a 20-second uncompressed talking-head AVI file at 320 by 240 pixels and 30 frames per second. We then encoded this clip using the RealNetworks and Microsoft encoders, holding the bit rate constant at 100 Kbps -- 85 Kbps allocated to video, 15 Kbps allocated to audio. We also compressed with QuickTime, but because we didn't have the Pro version of the Sorenson compressor, the results were not comparable. We used the de-interlace filter, and for Real encoding we selected two-pass encoding, a feature not supported in the Windows Media encoder. We then compared the results at full screen, and the RealNetworks clip did appear to be of slightly superior quality.

Quality can vary considerably, of course, depending on the type of content you're encoding, and the degree of motion, panning and zooming, for example. You should check out your content and draw your own conclusions. In addition, encoding time is greater with two-pass encoding, a factor that isn't noticeable with a 20-second clip but that could be very significant if you're encoding a content library. Don't expect Apple and Microsoft to sit still on this; in general, quality is improving across all platforms.

Our Editors' Choice award goes to the RealNetworks platform. RealNetworks offers a well-thought-out architecture with a rich feature set, and the quality of the implementation is superior to those of the others we tested.

However, unlike Apple and Microsoft, RealNetworks makes you pay for the fully functional versions of its software; wherever you see a product with "Plus" in the name, you know that's the fully functional version, and you'll need to ante up some cash. If you plan to use RealNetworks technology to serve many users, the costs can add up. And this is not a one-time expense: RealNetworks continually updates its technology, spitting out upgrades across its entire product line. You'll need to decide whether to go the maintenance route or use a periodic upgrade strategy. Either approach can mean significant expenditures: The 60-stream RealServer Plus starts at $1,995, and a server offering 400 concurrent streams can cost you more than $20,000. Upgrade and support run 40 percent of the total software license.

Although Microsoft Windows Media is a feature-rich streaming platform, the quality of this implementation does not meet the standards set by RealNetworks. Still, Microsoft offers some pretty good competition, and Windows Media supports a number of features that RealNetworks has yet to implement. For example, Microsoft includes a load-simulator utility, useful if you're launching a service meant to attract lots of eyeballs. Windows Media also enables screen capture as an encoding source, useful in the enterprise.

With Windows Media, you can say the price is right (it's free). The flip side of that is the uneasy feeling, which we got a couple of times during our tests, that the adage "you get what you pay for" applies. Yes, you can get everything to work, but it may take a fair amount of persistence and frustration.

The Apple QuickTime technology, as tested, is only adequate. The Darwin Streaming Server is not up to the standards set by Microsoft and RealNetworks when it comes to server configuration and management. And out of the box, PC encoding software from Apple for QuickTime content creation is very limited. For example, you can't do a live Webcast without a Macintosh and the Mac-only Sorenson Broadcaster software. Still, if you're a Mac shop with QuickTime authoring expertise and some third-party software -- a common combination for many media professionals -- you may want to consider QuickTime streaming: You'll be able to achieve high quality and full functionality, including live Webcasting.