RE: What does 95th %tile mean?
On Thu, Apr 19, 2001 at 05:18:02PM -0700, David Schwartz wrote:
Same algorithm, same raw data, different 95% answers, both valid, yet one is twice as large as the other. Great outcome for a billing system isn't it?
Any billing scheme based upon statistical sampling will, with some probability, err in the favor of one party or the other randomly. But it is important that the customer understands that he is being billed based upon statistical sampling and thus there are no "exact" measurements.
Fair billing sysems? The traffic "samples" are just the current value of a byte counter minus the value of that same counter <x> minutes ago, divided by <x>. If you wanted to be "fair", you would bill on exact number of bytes transfered. But on the subject of fair... Most people bill on counters from their customer aggregation switch, but if you transfer data between two customers you are billed as though you sent it around the world, how is this fair? If you send traffic out a paid transit or free peer it costs you the same, how is this fair? If you pay only the higher of your inbound or outbound traffic, and you are a web server pushing 100Mbit out, then you get that 99.9Mbit in if you pay for it or not, how is that fair? How many people actually count every byte they transfer, and how many ISPs are actually capable of matching your count with theirs so you know they are not inflating their numbers either deliberately or accidentally. Hell, do you think the price you pay is fair? There is a reason major ISPs dont publish their prices openly, and that is because if you are not helping some sales person make his Z8 payments or comparison shopping and demanding better prices, you are probably getting screwed. Some people will charge for putting you on redundant uplinks with HSRP/VRRP, some will not. Some people will have redundant diverse well-planned infrastructures and some will not. Some will charge for both directions and 95th percentile, some will charge for 1 and 90th percentile. They will all claim to have better networks then they really do in one form or another, and they will all hide behind an NDA to make sure you don't find out because like it or not everyone has some skeleton in their wiring closet. And yes some will have huge amounts of money yet be run by idiots, and some will have almost no money and scrape out a solid reliable network. That is the nature of business, and fair is what you are willing to pay for the service you get. If you don't like it, vote with your wallet. :P -- Richard A Steenbergen <ras@e-gerbil.net> http://www.e-gerbil.net/ras PGP Key ID: 0x138EA177 (67 29 D7 BC E8 18 3E DA B2 46 B3 D8 14 36 FE B6)
[ On Thursday, April 19, 2001 at 21:37:58 (-0400), Richard A. Steenbergen wrote: ]
Subject: RE: What does 95th %tile mean?
The traffic "samples" are just the current value of a byte counter minus the value of that same counter <x> minutes ago, divided by <x>. If you wanted to be "fair", you would bill on exact number of bytes transfered.
Well, it depends on what the customer is trying to buy. If all you're selling is average throughput then the customer would simply buy the smallest pipe necessary to transfer however many bytes per billing period that they need to transfer. However that's not generally what Internet customers want. Even a customer who can predict their usage wants to by N bytes per time interval of T, but if they have lots of valleys in their usage graph then they don't really want to pay for a full N/T sized pipe. Given a pricing scheme based totally on the maximum line rate of the pipe the customer will be forced to buy the fastest pipe closest to N/T that they can afford and make do with the delays this causes them. Similarly if all you're offering is a 10mbit or 100mbit pipe then the average customer is not not going to want to pay for a full 100mbit/s of Internet usage -- they'll never sustain that maximum for any significant period of time! So in both of those cases there's demand for some kind of peak usage based metering. Unfortunately true peak metering will almost always hit full line rate at least once during a billing period, so ISPs and customers must dicker around and find some fair Nth percentile value that they agree will be a "true" measure of peak bandwidth usage. Obviously there's also going to be some demand for other weird usage schemes too, such as off-peak usage. Perhaps some of these other schemes are where the Internet providers can learn a bit from power companies. As an ISP how much of a discount could you give to a customer who would be willing to have their pipe rate limited in direct proportion to your current upstream usage (i.e. so that their potential capacity was only made available when your upstream(s) have spare capacity)?
But on the subject of fair... Most people bill on counters from their customer aggregation switch, but if you transfer data between two customers you are billed as though you sent it around the world, how is this fair?
Well, there's the rub, isn't it! ;-) But there are ways around that.... It is possible to measure usage per customer on your upstream links only (and then work your other peer costs into the port charge, or some such). Rumour has it that the likes of Nortel have devised silicon that will count packets based on header values. On a smaller scale it's possible to use the likes of IP Filter or such on a border router to count packets to/from any given customer subnet. Obviously as a customer you'll want to know these details before you sign on the dotted line and you'll want to make sure they don't change over time too....
That is the nature of business, and fair is what you are willing to pay for the service you get. If you don't like it, vote with your wallet. :P
Isn't that the truth! :-) -- 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 (2)
-
Richard A. Steenbergen -
woods＠weird.com