Meeting the Demands of On-Demand

As viewers increasingly call on video-on-demand to watch their favorite films and programs, VOD vendors are hustling to meet their growing demands. As a result, the technology driving VOD is in a constant state of evolution, with vendors' cable and telco clients seeking new revenue streams.

Such advances are being driven both by developments in storage media and new-usage patterns for the on-demand platform. As more popular content becomes available through VOD, operators have to adjust their systems to provide a higher number of streams than they were originally designed to handle. Too much demand can become a crippling problem.

VOD functionality depends on three main components: the digital set-top client that receives the content, the network architecture that distributes it, and the headend video servers that store and stream it. Most vendors agree that the existing universe of two-way digital set-tops and Gigabit Ethernet-based, IP-centric cable and telco networks are adequate for meeting VOD's long-term needs. But as usage increases, cable operators and VOD vendors need to dramatically improve the headend equipment.

Technology used to deliver movies is well established, but advanced applications like on-demand advertising are driving changes in VOD streaming and storage.

“Our customers have rolled out movies on-demand, and they want to leverage that infrastructure to drive more revenues,” says Phil Simpson, product marketing manager for SeaChange International.

“Network DVR,” which stores programming at the cable-system headend, not on a set-top box, is an intriguing possibility. It would allow subscribers to time-shift all broadcast and cable fare. Though requiring massive storage and streaming, network DVR is the buzz of the business. Simpson calls it “one of the hottest applications out there right now.”

The business models for network DVR are still being tested. In fact, Cablevision, which planned to test network DVR this summer, is in litigation with content providers, which oppose the idea because they claim it's a violation of their copyright. But European telcos are already offering network-DVR functionality with local content, and most VOD vendors believe that such a service in the U.S. is inevitable.

Concurrent provides the VOD system for Time Warner Cable's new “Start Over” service, which lets viewers joining a program near its conclusion pop back to the beginning. That involvement has given Concurrent a feel for what's required for a more extensive network DVR. “[Time Warner] has 60 channels of content, and they're keeping it on the system for a couple hours,” says Tim Dodge, director of VOD marketing for Concurrent. “You need a system that can store and manage a huge amount of content. We've got systems out there with 50,000 streams, so we've got the experience to make the network-DVR model work when it comes.”

But scaling up a VOD system to store thousands of hours of content for network DVR remains a challenge. Early VOD systems were designed presuming that only 10% of subscribers would request VOD streams at any one time. But vendors say as more popular network fare is available on-demand, “take” rates could exceed 50%.

“We still have work to do figuring out how to scale in a real cost-effective manner,” says Mark Crandon, Kasenna VP of marketing. “A few years ago, customers were talking about storing 100-200 hours of content.” Now cable operators are expecting thousands of hours.

Creating an effective VOD model usually means finding a balance between memory and disk storage, often by storing content in different tiers. One of VOD's truisms is that a relatively small slice of content drives the vast majority of subscribers' usage. To address that, many operators have used distributed architectures that store the most popular content in disk caches at the edge of the network, while the bulk of content remains stored on a large centralized server.

A major VOD advance is the use of memory storage, either flash memory or dynamic RAM (DRAM). Memory-based storage has greater bandwidth and can provide a far greater number of streams than traditional disk arrays. But it can be anywhere from 50 to 75 times as expensive. “Some folks will say to do all memory-based streaming, and others say just disk-based streaming,” says Crandon. “It's fair to say that neither of those is the better choice.”

As VOD vendors roll out memory-based streaming, operators are looking to use memory storage for the most popular fare.

“Disks don't provide all the bandwidth that you need,” notes Joe Matarese, senior VP of advanced technology for C-COR, which is starting to incorporate memory storage into its VOD products. “So we're putting in DRAM as a more cost-effective way of getting extra streams.”

SeaChange's Simpson sees a mix of centralized and distributed storage architectures among his customers. Memory-based streaming, which SeaChange will begin shipping this fall, will provide further flexibility. Dynamic RAM (DDR2) memory costs about $350 per gigabyte, says the vendor, with flash costing $250 per gigabyte and SATA disk drives costing $4.50 per gigabyte. Thus, SeaChange estimates that memory-based storage is suitable for perhaps 1%-3% of assets.

“The most popular content will go in DRAM, while, in the middle, you have the medium-popularity content on flash drives,” Simpson explains. “The rest, the content you access once a week, you want to keep on disk drives.”