Monday, February 11, 2008

How Much Bandwidth is Enough?

Nobody yet knows how much Internet access bandwidth a typical user will need in the future, at peak times (average usage doesn't much matter). It is easier in many ways to model bandwidth requirements for entertainment video services. 

If a provider uses a "broadband" approach(in the sense of all linear channels being delivered to the user, whether or not the user is watching), it is a simple matter of ascertaining how many discrete video feeds one wishes to deliver, how much bandwidth each feed requires, and then doing some simple multiplication. 

 If one then decides to deliver all on-demand programming, one needs a switching infrastructure, and then must make some assumptions about simultaneous peak viewing. Will a typical user, at the peak viewing hour, want to watch one feed, two feeds or three, keeping in mind that one of the feeds might be recorded for later viewing while a second is actively watched. 

The Internet access portion of the planning exercise is more murky, but still hinges on video behavior. If business logic allows it, users might be able to stream video, even HDTV video, over IP connections. Whether this bandwidth is of the "public Internet" type or the "walled garden" type is less important, in some sense. 

 Assuming there is a revenue model, how much bandwidth must a service provider be able to provide? Whatever end users may think, a service provider will deliver bandwidth in amounts that allow it to make money, and no more. 

 So asking how much bandwidth users may want is probably less important than how much they are willing to pay to get that level of bandwidth. 

And so far, few users seem to have shown a willingness to spend hundreds of dollars to get symmetrical bandwidth, whether that is a T1 connection or a 50 Mbps symmetrical service from SureWest Communications. To use the old but useful analogy, all of us might enjoy driving a Lexus. But not all of us do. We solve our transportation problems, but not always with a Lexus. In principle bandwidth ultimately will represent that sort of choice as well. 

Just about anybody can buy a T1 connection today. But not all businesses do so, and few consumers do so. Granted, the bulk of consumer bandwidth requirements still will remain of the asymmetrical sort (barring a massive switch to peer-to-peer), so symmetrical bandwidth might not be the best analogy. Still, the question remains: how much bandwidth will consumers pay for? "Need" is it that sense a subsidiary question. 

There's no question typical consumers are showing a clear preference for paying more for higher bandwidth. The issue is the elasticity of that demand as service providers start to move into the "scores of megabits" range, and then contemplate bandwidths an order of magnitude higher than that (100 Mbps or more). 

 If one looks simply at the price-per-megabit, users have shown a wide willingness to pay $50 to $100 a month for unrestricted use of 200 Mbps to 500 Mbps of linear video (with implicit quality of service assurances). 

 They likewise have shown high willingness to pay $50 a month for a few megabits to several megabits per second of interactive Internet access bandwidth in the downstream direction, with no quality of service assurances. 

 Assume that most also have been willing to pay $50 a month or so for a wireline voice connection and you are looking at $150 to $200 worth of monthly revenue for services offering several hundred megabits-per-second of downstream bandwidth, plus services on top, using a highly asymmetrical network. 

That does not leave lots of headroom for networks that deliver more symmetrical bandwidth (scores of megabits per second in the upstream and hundreds of megabits per second for linear and on-demand video plus 100 Mbps for interactive applications). 

 In the consumer markets, the rule of thumb has been that $10 a month of incremental spending is a big deal. Still, shown a value proposition high enough, even $50 a month in incremental spending now has become fairly commonplace. 

So the issue might be more "how much will consumers pay?" rather than "how much bandwidth will they need?", as important as that question remains. 

There always are trade-offs engineers can make: bandwidth versus processing, processing versus storage, non-real-time versus real time, bandwidth versus image quality and so forth. Ultimately, consumers are going to drive access bandwidth with their wallets.

No comments:

Many Winners and Losers from Generative AI

Perhaps there is no contradiction between low historical total factor annual productivity gains and high expected generative artificial inte...