Pushpendra Mohta wrote:

The argument about billing based on destination being very expensive is not entirely true. Whether on not they bill on this data, most any one planning a global IP backbone measures traffic distributions very carefully, not only for network engineering but also to understand the cost of business. The granularity required to bill on this data may have to improve -but it is not a big leap to make.

Generally speaking, it is impossible, because aggregation removes information. I.e. the better aggregation is, the harder is to pinpoint the exact location of the destination. In other words, even destination-based charging (except very special cases like ISP in the middle of nowhere connected by one wire to the rest of the world) requires cooperation of many parties, since no party has any accurate cost-of-transport information locally. --vadim

pushp@CERF.NET said:

(rather elegant piece of text removed)

Why does then paying for many "information appliances" that tap into an "IP service pipe" in a similar fashion seem so outlandish ?

I don't disagree with much of what you have written, but playing devil's advocate I think it *seems* outlandish to some because: 1. Ostrich-with-head-in-sand change-resistance is rife. 2. From a superficial point of view, it seems that the costs are not usage sensitive. Tier 1 (whatever that mean) backbone's are building network as fast as they can get OC-12s deployed / hire engineering staff etc.; The planning process itself for (say) taking an OC-3 network to an OC-12 network is not a short one, let alone the deployment. If you have a small number of high bandwidth customers connected to one POP, your costs of connecting a few customers who are guaranteed (by SLA) to be able to get an DS-3 if they want it, but are billed at a rate of (say) a few Mb/s, is the same(*) cost as connecting customers billed for a flat rate DS-3. (*)=OK it may not be the same as usage based charging may well give you a higher port speed contention ratio, but the point here is with port speeds which are significant compared to backhaul speed, there is a significant risk cost in not deploying sufficient backhaul to cope with customer usage growth you can't predict due to not having a large enough statistical sample to work with. If you sell at flat rate, you can refuse or delay upgrades which would otherwise impare QoS to other customers. If you think about this in the context of any other good, so called usage tariffing is actually a relatively simple derivative product. If you are buying coal for a power station, you won't get a contract (or not at a decent price), where you can have however much coal you can physically get in trucks down the road at guaranteed price X, but you can instead order none at all. Instead, you might commit to buying Y amount of coal a month at price P per ton, have an option on amount Z more per month at price Q per ton (where P<Q). In a traditional market if you didn't misplanned, and didn't want X amount after all, you'd either have to store it (and incur costs), return it (and incur costs) or return it to the market at a possibly lower price (hence arbitrage). The situation is *not* entirely similar to electricity. If electricity was charged the way some providers charge per bit, I could order a power line and normally be charged for 300W of continuous consumption with no fixed charge. Then suddenly one day I could plug in a steelworks with no prior warning, eat many megawatts, and expect the same supply to work. If electricity was installed like this, you would have a huge fixed cost per month. If there were only a few homes per smaller substation, the fixed cost would be even greater (risk). However, neither ISPs nor consumers have sufficiently (yet) analysed the market to find out how this works. As you point out, every bit sent is likely to have a different cost, depending not only on destination, but on time of day too (think about the electricity example). If ISPs can't work out how to charge eachother per bit for peering arrangements (where, how, and to whom one offloads traffic in a peering context has substantial cost based implications) how can we expect them to not only develop realistic charging schemes for their customers, but also explain them to a relatively immature market? 3. Fixed cost charging is popular with customers. Many customers do not understand what they are buying. They do not understand the bandwidth (and thus cost) implications of (say) changing their web sites. They are frightened their bills might be astronomically increased by some activity on the line which isn't benefiting them proportionately (i.e. DoS attack, rogue employees etc. as opposed to extra web site hits, more employees finding useful information from the net, whatever). And the customer facing tools to analyse *why* bandwidth usage is what it is are, at best, primative. We don't yet have a ready market in programs to automatically recompress the .GIFs on your web site as JPEGs prior to transmission (or whatever) to save bandwidth costs. When the IT people (or whoever) attempt to get budget for their line internally, a fixed cost is more easily justified to the relevant bean-counter. An open-ended liability (which is what I've heard usage based charging referred to as) is unpopular. The people posting to NANOG 'disagreeing' with usage based charging are presumably merely voicing their dissatisfaction with it as a product offering. 4. The market will decide. Trite this may seem, but it's true. There are a lot of products which are half way between pure per bit charging and pure flat rate (you know this, CERFnet has at least one; I know this, I buy them from you :-) ), and these are essentially first hacks at a derivative market in bits (the way I see it). But just like derivatives, they aren't yet especially comprehensible. Awaiting cross-US bits to be listed with pork bellies... -- Alex Bligh GX Networks (formerly Xara Networks)