[ NNSquad ] Re: Fractional connections / duty cycles / usage caps

[Rahul Tongia <tongia@cmu.edu>]

There's a flip side to this that many people don't bring up.

If they want to charge me more for periods of congestion, they should charge me less for when there is idle capacity (assuming we are at equilibrium).  [today we might not be...]

Instead of charging less they could also let me use more (higher speeds) when there is capacity.  Shared media already have some of this going on but the question is one of fairness.  What happens when one or a few users "hog the bandwidth?"  I can bet there are fair, transparent mechanisms that one can devise (via, e.g., machine learning techniques) to only manage the usage of such users.  The catch is, of course, over time, all users tend to grow in their usage away from the previously predicted "norm."

By the same token, can't applications be made network aware? Not just "network status" per se but on a per link basis?  In the power world, people are striving for real-time power. This works because the bottleneck is mostly at the supply side (a little in transmission) (and most of the price implications, e.g., for peaking power, are at the supply end), but in the Internet world, many a times the choke points are in the last mile at the access side.  

One challenge with any system even if it could announce costs/prices for specific flows based on marginal costs would be that if I am a "power user" and use a mostly idle system (and thus can do it "cheaply"), it is no longer idle for others. The others may not even be doing p2p or other such activities that a power user might.  What we might need are applications that are aware of the network and able to wait for certain activities. Some applications are on timers (e.g., updates in the middle of the night), but we might want to think of real-time systems (including so-called p4p).  In the power world, companies are striving to move from Time of Use pricing (for big users) to real-time pricing.  Easier said than done.

Rahul

On Tue, Sep 1, 2009 at 3:34 AM, Wes Felter <wesley@felter.org> wrote:

I'd like to distinguish between three cases:

A) You buy a connection with X% duty cycle and the ISP prevents you from sending/receiving more than that. Your bill is the same every month.

B) You buy a connection with Y GB/month usage cap, but the ISP allows you to exceed it and then charges you (a lot) more money.

C) You buy a connection with Y GB/month usage cap, but the ISP allows you to exceed it and then cancels your account if you do.

(I assume the reader understands that a duty cycle and a usage cap are the same thing, just expressed in different ways.)

I think A is good but B and C are bad. (I think all three options should be legal, but I'm not talking about legalities in this message.) I *want* a bursty Internet connection because my usage is bursty, but I also want some kind of "safety valve" that keeps my bill under control. In summary, usage caps are good *when implemented properly*, but *virtually all broadband ISPs are implementing usage caps wrong*. When an ISP says "your bill will be somewhere between $50 and $150" the resulting customer outrage is entirely predictable and IMO justified.

Wes Felter