Networked-Versus-Integrated PVR Debate

Serendipitous surprises, like the unanticipated run on Home Box Office's subscription video-on-demand test in South Carolina a few weeks back, sometimes spawn all sorts of interesting developments.

As the story goes, HBO dangled on-demand viewing of Sex and the City
and The Sopranos
to 26,000 Time Warner customers in Columbia, S.C. About 3,000 HBO customers took the bait, for an 11.5-percent response rate. That alone is pretty amazing. But there's more: It all happened in less than two hours.

What happened next is the very reason these things are called tests: A frightful anxiety attack struck the headend equipment that sets up streaming sessions with set-tops. Way more people had arrived at the party than the headend controller ever anticipated, and it went into a tizzy, speaking in tongues. The server, hearing unintelligible babble from the controller, tuned out. Time Warner downshifted, startled, but well aware of the sweetness of the problem.

Whether you believe in subscription VOD, or in personal video recorders (PVRs, like TiVo), the effect on subscribers is essentially the same: They get to watch TV when it suits them. The effect of on-demand TV on cable technology is not the same, though. It requires a decision: Should content be stored on the network, or on a hard disk at the home?

The question of where to put the bucket is a chewy technical debate that goes way, way back.

Consider this: Research data indicates that within two weeks of buying a PVR, users shift about 60 percent of their TV viewing to on-demand. (It's certainly true here in the geekosphere: Weather is the only live stuff we watch. Everything else is stored for empty evenings.) Modeling sessions and streams for that level of anticipated usage is a mind-bender, as any of the bleary-eyed souls who do bandwidth planning know.

Cable, unlike other broadcast delivery formats, is the one group that could go either way. It has enough brute force bandwidth to keep the storage bucket in the headend: A 750-MHz cable system translates into some 4.3 Gigabits per second (Gbps) of downstream (headend to home) capacity, if all channels were digitized and compressed before delivery.

It also has 10 million-plus digital-video subscribers. The installed digital base doesn't innately do personal video recording — TiVo wasn't yet a twinkle in digital's fertile eye, back when cable's digital boxes were invented. Still, the deployed boxes generally do have the right plug on the back panel to attach a sidecar PVR. And, integrated set-top/PVR boxes are on the way, perhaps by the end of this year.

What to do? Cable technologists and marketers suddenly find themselves amid a complicated economic analysis of centralized versus distributed storage.

On the one side, the PVR proponents: TiVo, Replay and the set-top suppliers themselves. Build storage into digital set-tops, they say. Or, strap it on as a sidecar. (Motivation: Sell more boxes and hard drives.)

On the other side, the networked PVR proponents: The nCUBEs and Concurrents of the VOD world. Centralize storage in the network, they say. Deliver it as needed. The bandwidth is there. (Motivation: Sell more servers.)

And then there's what the two sides say about each other. Networked PVR proponents believe set-top storage to be cost overkill: Spending $200 more to snap a hard drive into a $230 box is nuts, they say. Centralizing storage into the network (headend or hub) means more people can share more storage, thus spreading cost.

Plus, say the nCUBEs and Concurrents, putting a device as inherently failure-prone as the hard disk drive inside
the home introduces ugly service problems. When the PVR's hard drive fails, who absorbs the venomous call from the subscriber whose entire family's favorite TV programs were destroyed? The box manufacturer, or the cable operator who installed it? By striping the TV content such that it splays across several remote servers, nothing is ever lost, they say.

PVR's counter: Initiating real-time, bi-directional sessions between a digital box and a remote server is new, and tricky. (Witness Columbia, S.C.) The number of sessions per second that can be handled by a headend controller may never be high enough for on-demand TV. And, PVR people say, this isn't like modeling for movies on demand. Usage patterns for time-shifted TV are alarmingly higher than for time-shifted movies. Consequently, storage and bandwidth needs will be alarmingly higher, as well.

There is no tidy answer to the question of centralized versus distributed storage. Both camps make plausible points. Meanwhile, Columbia proved that subscribers go for TV viewing convenience. More and more program networks will emulate HBO's maneuvers. DirecTV and EchoStar already offer PVR. Probably the best way to find the answer is to avoid analysis paralysis, and jump in: Sometimes, the most useful knowledge comes from doing.