An Open Letter To Netflix

by Brian McConnell

I've been using Netflix since October 2000, so I think that qualifies me as an early adopter. I love the idea, and have watched hundreds of movies and documentaries thanks to their innovative service. I haven't set foot in a video store in several years.

Until recently, I had been watching 5 or 10 movies per month, sometimes more, sometimes less. That is, until I got a high definition TIVO for my DirecTV system.

Since then I have watched only a handful of DVDs. The backlog of videos on my TIVO list is lengthy, and grows moreso every day. I still watch Netflix DVDs, but much less frequently, and usually when I am traveling.

It's not that I find Netflix inferior, I still love it, it's just that the TIVO records so much stuff, much of it quite watchable, that I don't often have time left to watch DVDs. While people might have enough cash to pay for both, they only have so much time.

DVDs are great, but satellite television is in many respects a better delivery mechanism. A Netflix + satellite PVR combination would be a potent combination indeed. The satellite system provides the fat pipe needed to deliver content, the PVR provides the storage medium.

Netflix and TIVO are allegedly working on a streaming media service that will allow people to download videos on demand via broadband. Nice idea, but also a perilous one. I'll explain the reasons later in this article.

Netflix is based on the concept of delayed gratification (movies usually sit in my queue for weeks, sometimes months, before they arrive, as I have such a huge list). So instant delivery is hardly necessary.

This can be improved even further by partnering with existing content providers and PVRs to make the Netflix queue a TIVO feature. Add a title to your Netflix queue and it is automatically beamed into your TIVO, which grabs it from the first available source. Netflix provides the queue management (which is the real value they provide). HBO, Directv or whoever provides the content delivery. Simple, and this concept can be applied to any PVR enabled delivery channel (satellite, cable or broadcast).

Netflix, via periodic dial-up sessions initiated by the PVR, would know which films had been transmitted to your PVR, which ones you'd watched, and how you'd rated them with the thumbs up/thumbs down buttons. Netflix would therefore know which films had not been sent, and would use snail mail as a fallback delivery option.

Customers would presumably have the option of setting delivery preferences like:

  • Send all DVDs via snail mail, regardless of PVR status

  • Only send DVDs via snail mail if not received by PVR within X days

  • Automatically delete queue items when transmitted to PVR

This type of arrangement would change the context of usage. For at home viewing, customers would largely rely on queues videos downlinked to their PVRs, except for uncommon titles that cannot be delivered in a timely manner. The DVDs would become the take-out option, which is exactly how I use Netflix today. I watch TV at home using my TIVO PVR. I stuff the DVDs in my laptop bag for business trips.

Soon, I will not even need Netflix for take-out, as TIVO is rolling out new features to export recordings to PCs and to burn DVDs. Once the DVD burn option becomes available for Directv, I doubt I will use Netflix even for travel.

Video Almost On Demand Versus Video Via Broadband

Video on Demand has been the Holy Grail for telephone and media companies for well over a decade. It is a deceptively simple concept, but full of technical obstacles and hidden expenses. Video Almost On Demand, by contrast, is cheap, and in the case of a service like Netflix, can be done quite simply.

The problem with video on demand is two-fold. First, it is a bandwidth hog, requiring several megabits per second for standard resolution video, and closer to ten megabits for DVD quality video, and still more for HD quality video (although H.264 will deliver HD quality video at about 8-10 megabits). Delivering high quality video in real time via broadband is not feasible except over VDSL or fiber optic lines. In theory, cable modem service can deliver enough speed, but cable is only fast because it takes advantage of the bursty nature of Internet usage. If a lot of users start downloading multimegabit streams for hours at a time, the performance and economics of cable Internet service go out the window. Solving this problem requires finer network segmentation with edge servers deployed throughout the network to reduce the redundant transmission of data. Ultimately, this is a solvable problem, but the solution is costly, and it is questionable if people really need instant on-demand access to everything.

The second problem is format conversion and storage. Converting tens of thousands of DVDs and digital television broadcasts to a PC friendly format such as Quicktime or MPEG4 is time consuming and costly. While the actual disk storage is not so expensive, video compression takes both human and CPU time. One could leave the recordings in their native MPEG2 format, but that would increase transmission bandwidth requirements, which is a deal breaker for virtually all broadband users, few of whom have downlink speeds faster than 3 megabits. While MPEG2 bitrates can be dialed down, this has a direct, and negative, effect on video quality. People might tolerate blocky video in a Fox News broadcast. They won't appreciate this when watching a movie on a high resolution TV.

Video Almost On Demand, by contrast, can be done very cheaply and using existing hardware. So let's look at a simple solution that can be implemented entirely using existing hardware and satellite/CATV delivery systems.

The TIVO box would be modified to engage in a simple dialogue with Netflix. This dialogue would proceed as follows:

1) User authentication
2) Request recordset of titles currently in Netflix queue
3) TIVO box compares queue recordset against its to-do list, ignores titles already in internal queue
4) TIVO box attempts to add new titles to its to-do list
5) Reports success/failure to add new titles to internal to-do list back to Netflix
6) Reports user ranking for recently viewed titles, which are automatically removed from the Netflix queue

Note that this is a simple transaction, essentially no more than a login followed by a series of simple database queries. This will work on any PVR, regardless of transport media (the more channels available, the greater the odds of prompt delivery). I would estimate that once a specification is defined, the time required for software development would be pretty minimal on both ends.

The business case for doing this is pretty compelling. Reducing dependence on postal delivery reduces operating costs substantially, especially to "power users" who view a disproportionately high number of titles and are more likely to own a PVR. Piggybacking on existing content delivery systems (e.g. CATV, satellite) eliminates the need to develop a separate and expensive video over broadband system. Postal delivery becomes an fallback for less common titles, and for users who specifically want DVDs for take-out viewing.

Lastly, because this feature is so cheap to implement, not doing so poses substantial risks, because what is easy for Netflix to do will be equally easy for its competitors to do. Blockbuster and Amazon are both making forays into this space. Netflix deserves credit, and a prosperous future, for pioneering this concept. So they will do well to follow the motto "deal with change before it deals with you" because cheap and simple solutions like this have a way of upending the most carefully calibrated business plans.


2005-05-09 08:50:27
Broadcast Disks
This article prompted the memory of a concept that one of my professors (Stan Zdonik) was working on 10 years ago, dubbed "Broadcast Disks". (One frequently cited paper is "Are Disks in the Air" Just Pie in the Sky?.

This broadcast disks work is basically an attempt to solve the access scheduling problem, based on a hierarchy of delivery mechanisms. This could be the USPS truck used by Netflix (high latency, high bandwidth), cable/satellite delivery (medium), or potentially even broadband (low).

And it's interesting: the spinning analogy developed for Broadcast Disks can hold even for the truck, where presumably there has been a specific request. Maybe I couldn't see it then, but I have a shared account with my wife; or maybe I wanted to see that movie again. Either way, it gets rescheduled. A larger cache of course would help!