Internet Buzz with Richard Wiggins | 9
|
|
||
| Volume 1, Number 27 | July 9, 1998 | Internet Buzz main page |
|
|
||
Will You Pay for Bandwidth? The Berkeley INDEX Experiment |
  
|
|
Dr. Varian, back in 1993, you and your colleague at Michigan, Jeff Mackie-Mason, advanced a simple argument: that scarce Internet bandwidth would be most efficiently allocated if users paid more to get their data through at times of congestion. You argued that when the Net was busy, real-time multimedia users might pay substantially, but that the cost of some services, such as basic e-mail, would approach zero. I remember well the furor your ideas unleashed among some people, who declared "the Information Highway must be free!" Now, we seem to have settled on $21 per month as a standard number for basic Internet access, and, as you predicted, there are a number of free e-mail services on the Net. And it seems we don't hear so many utopian calls for free access for everyone. Are we ready to take the next step, in which end users pay for quality of service? I think that this is inevitable, especially as more and more people start using Internet telephony, video conferencing, and related real-time technology. You and colleagues at Berkeley have launched a project, known as INDEX (for Internet Demand Experiment) that seeks to measure real users' willingness to pay for quality of service. How much engineering was necessary to implement the INDEX testbed? The engineering was mostly done by Richard Edell, a very talented graduate student in EECS [Electrical Engineering and Computer Science]. We did an early prototype in March 96, and it was up and running by March 98. Richard probably spent about half of his time on this project during that period. As with most projects, a lot of the time was spent in getting funding, lining up sponsors, doing the logistics, debugging and the system integration. Do you have some preliminary results based on your surveys of real users' attitudes towards paying for quality of service? We're getting very clean data from the experiment. One of the most striking features is how sensitive our users are to priced service: they're definitely interested in conserving bandwidth when they have to pay for it! Isn't it a lot easier to implement a QoS feature -- including billing -- in a testbed than in the real world, where a dozen or more companies may help route packets? Absolutely. If this is to scale there has to be some standard for signaling QoS. I like Dave Clarke's model with a single bit being used to signal in-profile and out-of-profile traffic.How will settlements occur? I assume I won't pay an Internet bill with itemized fees for every company that carries one of my packets! There are fewer backbone companies every day, so maybe it won't be so bad! More seriously, the post office manages settlements without this problem and I expect that the whole process will be transparent to the user. Recently I heard a Bell Labs scientist joke that eventually every packet sent on the Internet will put itself up for auction, bidding to find carriers to deliver it. Are we headed for that sort of world? I could easily see this happening at the flow level. After all, markets are about the best mechanism we know for managing complex, real-time distributed systems. |
|
|
Comments are welcome
Produced by Rich Wiggins and
All Rights Reserved. Legal Notices.
Created: July 9, 1998
Revised: July 9, 1998
URL: http://webreference.com/outlook/column27/page2.html
