On Thu, 19 April 2001, Geoff Huston wrote:
It leads to the situation where the provider confidently asserts that the 95% value was xkbps, and the customer confidently asserting ykbps and both readings are equally valid, with both measurements using the _same_ measurement technique. How is the consequent billing dispute resolved _fairly_?
The great thing about changing jobs every once in a while is I get to experience the joy of looking at things from the outside. Because colo's use a lot of power, I'm learning a lot about how the electric companies work. Geez, if you thought telephone companies were bad :-) The power industry has a version of the 95% billing. Like ISPs, originally it was a simple measurement every 5 (10, 15, etc) minutes. Figure out the peak measurement, and you are done. Users figured out how to better control their usage, i.e. turning on motors for 9 minutes and then turning them off for six minutes. Were there billing disputes, you betcha! Now electric companies use a variety of sliding windows, and different ways to calculate usage. The top-of-the-line power meters have over 100 different ways to measure (and therefor different ways a utility can charge) power consumption. Each becoming progressively more complex, trying to eliminate ways for people to game the system. To resolve some billing disputes, the top meters include "pulse" connections so you can precisely synchronize your check meter with the utility's meter for auditing. So, does someone need to develop an Internet meter for your Internet line with a dial which spins faster when you use more packets?
To resolve some billing disputes, the top meters include "pulse" connections so you can precisely synchronize your check meter with the utility's meter for auditing.
Exactly! The issue is that 95% systems produce different answers on the same data set when using the same algorithm simply by altering the relative phase of the two 95% measurement systems. The only way to get the two systems to produce the same answer is to get them to synchronize their measurements in time. I like the idea of the visible 'pulse' to allow the customer to sync with the provider. Unfortunately I'm not sure I can see what the IP meter pulse equivalent would be!
[ On , April 19, 2001 at 16:42:38 (-0700), Sean Donelan wrote: ]
Subject: Re: What does 95th %tile mean?
The power industry has a version of the 95% billing. Like ISPs, originally it was a simple measurement every 5 (10, 15, etc) minutes. Figure out the peak measurement, and you are done. Users figured out how to better control their usage, i.e. turning on motors for 9 minutes and then turning them off for six minutes.
Hmmm... I've never seen that type of power measurement in household, small business, or farm usage; at least not anywhere across Canada. Here the meters are the type that spin at a rate determined by the current flow. I could check my texts on the subject but IIRC the meter actually counts up the total current flow through the meter regardless of "rate" and then the monthly reading is divided by the number of hours in the month to get watt-hours and that's what we're billed for. I.e. it's done almost exactly the way bulk throughput Internet usage is billed, but since a power meter can't count watts as directly as a router counts octets they have to measure the rate and derive the count. The meter has to be a fairly precise and carefully designed instrument though because it must react quickly enough to current flow changes so as not to be fooled by a quickly pulsing load. The only tricks to help reduce peak-usage that I've seen around here are the likes of electric water heaters that are diabled during peak load times by a little relay out on the street. Instead of using a separate meter they simply bypassed the meter and charged a flat monthly fee people figured out how to hook things up to their water heaters illegally, etc., etc., etc. There's not much of that around here any more because of that and partly because the pricing structure they originally used for this scheme was flawed: it was based on some statistical average usage for the customer base. Trouble is the customer base changed and changed their usage patterns more dramatically than they accounted for and their contracts locked them down in what turned out to be inappropriate ways. In the long run it would have been better to fork out the capital costs and use two meters with different rates for the different types of usage (i.e. have a lower rate on peak limited controlled usage). I wonder too if the usage patterns didn't change enough that there was no more reduction of peak loads when the water heaters were cut off, what with fewer housewives doing laundry, more use of cold water for laundry, etc. In any case the analogy with power meters is entirely flawed because Internet usage on a fixed rate pipe is an entirely different critter. In effect the custoemr is buying voltage, not current, and that's why a 95'th percentile metering of the peak "voltage" is more appropriate for Internet billing in many cases. As a customer I can download the same movie at full rate over a short period of time (eg. the duration of the movie) or at a very low average rate spread over a whole month. Obviously if I do it the first way then my ISP want's to bill me for using the full rate (even if I only did this once in the month), but if I do the latter then the ISP only needs to bill me at the very low average rate because that's all I used. Either way my total byte count is identical. However unless I'm extremely patient I'm not going to wait a full month to view the movie, no matter how little my bill might be at the end of the month. Nth percentile billing works best when ISP hands the customer a rather fast pipe, eg. a full-duplex Ethernet interface, but when the customer's usage will not keep that pipe full. The customer wants the fast pipe, but does not want to pay for it as if it were full all the time. The important factor from the ISP's perpective is not the total throughput (though that's definitely not to be ignored), but rather the peak usage. So since the ISP can often save money by handing low-cost high-speed ports (eg. etherent) to the customer, he still wants to get paid for peak usage. The value of `N' in an Nth percential billing agreement is set to balance between the customer's perceptions of his requirements and the ISP's very real need to buy upstream bandwidth that might just possibly be able to fill the customer's pipe on demand. Oh, and obviously this is another reason why the ISP will also charge a "port charge" and why they might include a certain base rate (eg. 1mbit) in with the port charge. -- Greg A. Woods +1 416 218-0098 VE3TCP <gwoods@acm.org> <woods@robohack.ca> Planix, Inc. <woods@planix.com>; Secrets of the Weird <woods@weird.com>
participants (3)
-
Geoff Huston -
Sean Donelan -
woods＠weird.com