Showtime for Streaming Video
NewMedia January 1999
by Matt McMakin, Samuel Murphy and Jeff Sauer
John Glenn's latest blast-off may have been the first HDTV broadcast, but more people watched that event over the Internet than on HDTV sets. This signals that if there's something gripping or important being streamed -- whether it's presidential scandals or training seminars -- people will watch. In this two-part article, the NewMedia Lab takes a careful look at Internet and intranet streaming video servers.
First, we examine the Internet options. This once-crowded field of startups and specialty codecs has come down to two combatants. RealNetworks, with a huge installed base of RealAudio and RealVideo users, is readying RealSystem G2, a major release. Microsoft, on the other hand, is the 800-pound gorilla that has come a long way with NetShow 3.0.
In the midst of all this, Apple's QuickTime is the dark horse. Even without a server, downloaded QuickTime clips and pseudo-streaming FastStart for QuickTime video have allowed the masses to tap into Web video; and it's rumored that the next version of QuickTime will have a streaming server. Better quality and smoother performance have so far beaten convenience of streaming (see sidebar, 'Gimme a Side Order of Video'). Yet, Internet bandwidths are growing with increasing numbers of 56Kbps modems, cable modems, and LAN connections, and trends show streaming steaming ahead.
On the other side of the wall, the much greater bandwidth of corporate LANs opens the door to very high-quality video. While Microsoft and RealNetworks may ultimately control the high-bit rate market as well, several other companies have staked a claim. Cisco Systems, 3CX, InfoValue, Vsoft, and Xing Technologies all have intranet video servers, many of which have specific features for corporate applications. One other longtime intranet player, Starlight Networks, was purchased by PictureTel just prior to our testing. PictureTel will retain Starlight as a separate business unit, but initial uncertainty about the future of Starlight's StarCase and StarLive precluded our reviewing them at this time. I-Cast was also recently purchased by FVC.com, who will continue to sell the I-Cast line of intranet broadcast products.
Optivision also makes an MPEG encoding "black box" that you can hook up to the network. This box offers only bare-bones streaming broadcasting of either live or stored content for real-time news or training purposes. However, you don't get full administration features. Minerva Systems and Zulu Broadcasting are working on similar boxes.
The View from the Top
While all of these products enable "streaming video," the difference in bandwidth means two entirely different problems. In best-case scenarios, 28.8 and 56kbps modems have less than 5% the bandwidth of 1x CD-ROM drives (remember those 1x CD-ROM drives?) This mandates heavy-duty compromises in video quality and frame rate.
Microsoft and RealNetworks use proprietary low data-rate codecs that try to accommodate the unpredictability of the Internet, which often operates well below peak rates due to congestion and other irregularities. Both companies have also made efforts to dynamically match available bandwidth to provide the best quality at any given width.
However, even with these advanced new codecs, video quality at modem speeds is, shall we say, rugged -- always with visible artifacts and often leaving out the detail. Simple talking heads can be effective at 28.8, though you're lucky if you can get more than a couple frames per second. Running video at resolutions over 160-by-120 can force it into a virtual slideshow. 56Kbps modems improve frame rates and picture quality, though images remain soft.
Fortunately, audio over the Internet can be quite effective, and good audio greatly improves the impact of even marginal video. Low data-rate audio codecs have moved beyond just being very cool technology to making streaming audio quite palatable. Music sites offer samples to entice purchases, and radio stations regularly stream programming, including news, sports, and music.
Moving over to LANs, 10BaseT Ethernet has 200 to 400 times the bandwidth of most modem connections (or, about an 8x CD-ROM) and room for multiple simultaneous streams of very good quality MPEG-1 video. 100Mbit infrastructures make it practical to think about broadcast quality MPEG-2 video. The intranet servers we tested are all MPEG-1 capable, and most support MPEG-2 as well. Ideally, servers should be codec-agnostic and able to stream any video file.
The View from the Network
Talk to network administrators about streaming video, and chances are most will voice concerns about overloading their network. To assuage these anxieties, several of the server products we cover include controls for limiting streams or pruning them back when other critical network services, like e-mail, require more bandwidth (see features chart for more details).
Virtually all servers use a variation of UDP (User Datagram Protocol), the base protocol for streaming data packets. UDP limits the extra overhead of packet monitoring and error correction, which is wasted on media packets, since re-sent packets would arrive too late anyway. Unfortunately, many firewalls are configured to shut out UDP packets, since their content is not easily identifiable. Both RealSystem G2 and NetShow support TCP or HTTP streaming as alternatives, even though the overhead of these protocols is a drain on performance.
Currently, most Web video is unicast, serving a separate video stream to each viewer. For high-demand live content, unicasting can be a big drain on server resources. IP Multicast sends out a single stream and lets routers replicate it, potentially far away from the server. Unfortunately, while virtually all new routers and switchers include multicast capabilities, most default configurations don't have the feature enabled. Worse, existing ISP and corporate equipment is often too old to support multicast. Nonetheless, enabling multicast is a necessary precursor to widespread consumer acceptance of live streaming video.
Two Roads to Multimedia
Like adding breadcrumbs to make meatloaf, you can get more impact out of audio and video by stretching it with linked Web pages and graphics. Both Microsoft and RealNetworks provide ways to enhance media streams and create synchronized presentations with HTML-like scripting. RealNetworks uses the SMIL format, a W3C standard, while Microsoft uses its own ASF and ASX formats.
Microsoft's ASF format wraps both media and metadata together in a single file. While ASF fits into Microsoft's DirectX media architecture, it's a less flexible way to create content. To edit ASF metadata, you have to strip off the header file, edit it, and then recompile it with the media. Editing the media itself can disrupt positioning of the related metadata; e.g., shortening a video file can throw off the timing of URL flips (requests for Web pages) or looped playback. Another minor complication: To allow users to access an ASF file off a Web page, you need to program an ASX file, which tells the browser to call the Windows Media Player and load up an ASF file. This means that you have to keep track of yet another file. RealNetworks has latched onto a W3 standard, SMIL, for synchronizing video with other media. SMIL files can control other RealPlayer-supported events like text and transition effects, Flash animations, timing (with SMPTE timecode support), plus URL flips in a browser. A stand-alone SMIL text file references media and contains all metadata and control information. Separating the media from the metadata makes it easier to create templates, edit media, and control other components.
Microsoft NetShow
With NetShow 3.0, Microsoft continues its increasingly scrutinized practice of integrating capabilities into the operating system. No longer a free-standing product, NetShow is now a system service, available as a free download for NT Server 4.0, and something that will be built into Windows 2000 Server. When enabled, system services automatically load on boot, allowing any Windows NT Server to stream video‹no multi-client NT Server license is required. And free is a compelling price, especially for those just beginning to explore streaming video.
With NetShow's administrative tools, you definitely get more than you pay for, including some features that aren't available from RealNetworks. As a whole, NetShow does a good job centralizing administrative tasks so that you can monitor everything that¹s going on. RealNetworks' G2 is more of a free-for-all environment, where the people who create content also publish it on the server, and the server has an open port that will stream whatever is published -- akin to traditional Web serving. NetShow can also work this way for video-on-demand files. But, if you're talking about live or scheduled broadcasting, you have to register users and specify what they can do on the server, which lets you keep close tabs on broadcasts. And if you're already running NT and IIS, you¹ll be familiar with NetShow's security and organization model.
NetShow's client is, simply, the built-in Windows' Media Player, making any new Windows system a preconfigured client. However, only the most recent Media Player 5.0 can play back NetShow 3.0 streams, so most will have to download an upgrade. Taking a cue from RealPlayer, the new Media Player adds "favorite" streaming video sites, author and copyright information, and menu-accessible file statistics, but is essentially a low-frills application with basic VCR-style controls. With RealSystem G2 still in beta and awaiting a Mac player, Microsoft is alone in supporting Mac clients. While it should get credit for the effort, we could not play even local files with the beta version posted during our testing.
NetShow's encoder creates ASF files and is also free, but in this case you do get what you pay for, since it has a relatively limited feature set. The process is also a bit of a pain since the encoder always recompresses video data regardless of what codec was used originally. (You can compress directly to ASF using the low-end Osprey card, but ideally you should be able to capture ASF within a program like Premiere using a wide range of capture cards.)
Then, to create interactive presentations, you bring the ASF file into a stripped-down version of Digital Renaissance's T.A.G. Author, which is also bundled with NetShow. Here, you use a crude timeline to synchronize your video/audio with URL flips or other events. Alternatively, you can bring AVI files into Sonic Foundry's Sound Forge, editing your audio and video, adding ASF tags in its more sophisticated environment, and export ASF files. However, Sound Forge cannot edit existing ASF files directly (convoluted workarounds are possible).
NetShow supports an encoding range of 2.4Kbps for mono audio up through 3Mbps for high-quality video. ASF codecs include Truemotion from Duck, VDO, H.263, but Microsoft focuses on what it confusingly calls "MPEG-4." An MPEG-4 standard has not yet been ratified by the MPEG committee, so Microsoft¹s implementation may very well differ from the official standard. To top it off, NetShow does not support standard MPEG-1 or MPEG-2.
To compensate for fluctuating bandwidth, NetShow lets you include two different bit rates in a single source file. You choose two video quality levels‹say, 56Kbps and 28.8Kbps -- and, in theory, the client gets the higher or lower stream depending on what's available to them. Microsoft calls this Intelligent Streaming. According to our tests, however, Intelligent Streaming is not that smart: It doesn¹t support audio, and it fell well shy of RealNetworks' SureStream in performance and data-rate flexibility. |
