4 ms·

No, I am not arguing that bandwidth is unlimited. That would be prima facie impossible. I'm somewhat confused and distressed to read a summation of my argument

No, I am not arguing that bandwidth is unlimited. That would be prima facie impossible. I'm somewhat confused and distressed to read a summation of my argument which implies such an absurdity; it seems objectionable in the context of a calm and reasoned debate. I'm sure that you meant no offense. I'm not incapable of arguing absurdities, but I hope that this is rare, or at least I hope that you may assume so for courtesy's sake. Bandwidth is not unlimited, but the point seems irrelevant: costs do not scale linearly, in any sense, for any participant. It does not cost the ISP more or less if my modem is provisioned for 10 or 100 Mbps. It may or may not cost them more money if that bandwidth is actually used, but very few individuals have any large impact on their ISP's operating costs. The aggregate utilization does affect this, but the marginal operating cost per customer is much closer to "fixed/negligible" than "linear in proportion to usage". Correspondingly, customers are not billed based on usage (in the normal case). The implicit argument is certainly that the price should be lower, but they did not argue a linear price relationship. There is no reason to believe their argument would be invalidated if that were the case, but they did not.

greggyb 8y ago
When an ISP leases bandwidth to a customer, they must reserve some amount out of their total available. They may reserve the full amount they have leased, or they may reserve some portion of it. Either way, the reserve is finite and proportional to the actual total bandwidth (the proportion may be 1:1, which is the simplest, but the proportion does not matter for this discussion). The marginal cost of provisioning me 100Mbps is therefore not limited to the cost of the hardware and setup, but is the opportunity cost of reserving bandwidth for my use. They may not lease this reserve to anyone else. This cost is reflected in the price. Additionally, we can observe that ISPs tend to grow. Even in a fixed area, they will upgrade hardware over time to increase total available bandwidth. This is driven by demand by existing customers for more bandwidth and new customers. This hardware is very expensive to provision. In fact, I expect that going from total bandwidth x to 2x likely costs more than twice as much as x. I would also argue that it likely costs at least thrice as much. Products of high fixed cost and finite supply will never be sold at marginal cost (or at least not for anything approaching profit). They will approach marginal cost as N increases (where N is the size of the finite supply). Another way we can think of a customer leasing bandwidth is that they are purchasing a share of that large and expensive network equipment. I posit that these are valid ways to look at it. The first, because there is not infinite bandwidth, ISPs must price it like the scarce good it is - prices are how we address scarcity. The second, because ISPs must not price their service as if they are never making another network upgrade. Again, I am willing to haggle on the cost of 2x performance. I would make all of these arguments for any value under 3x price, as I mentioned above, which is clearly stated in the original parent. There is no need to stick to a strict linear interpretation, and I have said as much more than once now.
- sethrin 8y ago
  For residential service, there is no reason to expect a 1:1 reservation. This would be extremely wasteful. The exact number is not relevant, but residential bandwidth is oversold as a matter of course. Talking about capex and aggregate costs also seems to be drifting from the main theme here: whatever it costs to double the network's capacity is going to be pretty circumstantial, and not directly relevant to the topic of consumer pricing. <i>Ceteris paribus</i>, the marginal cost per individual is still negligible, and the price per unit of provisioned bandwidth only loosely coupled to cost. ISPs charge extra for higher-tier packages because having only one tier of service is leaving money on the table, and because they can.
  - greggyb 8y ago
    You're right. The reservation ratio doesn't matter - we agree on this and have both stated it's irrelevant. We're talking about average internet performance in a nation. This entire thread was kicked off by a comment about Akamai indicating the US as a whole has faster broadband than e.g. Germany. The cost to wire up the US with an average speed of 2x that of Germany is absolutely an influencing factor in the cost of bandwidth to a consumer. We are talking about performance, not widgets. To double the performance of a nation's network from p to 2p, it is not unreasonable to expect that the cost is greater than double or greater than 3x the cost of performance p. But, if you want to proceed from an analysis of marginal cost, we can do that as well. ISPs do not fit the model you are using. Marginal cost for something like an ISP is not a single number. There may be 10,000 consumers with a marginal cost near 0. Then customer 10,001 may have a marginal cost that is "install a new backhaul to the network backbone and extra 10G or 100G routing equipment to serve your neighborhood". Obviously customer 10,001 doesn't get stuck with a bill of $100Ks to be added to the network. You are attempting to pigeonhole an ISP into a model that assumes a single, flat fixed cost and (effectively) unlimited production of a good at a low marginal cost. That is not the industry we are in. The fixed costs of an ISP are a stepwise function, not a constant. You cannot model an ISP effectively with an assumption of constant fixed costs. All other things being equal, the marginal cost of a given customer is <cost of new network equipment or major network upgrade> / N, where N is the number of customers that can be served before the ISP has to install new routing hardware, or in the unhappy path lay new wire. Now, it is obvious that I have simplified things above. ISPs do not build infra for N customers, then wait for customer N+1 to build out the infra for the next N customers. ISPs invest in their infra in long planning cycles. They cannot effectively build out in this manner in response to a single customer. Thus, for years at a time, the supply of bandwidth is effectively fixed. With a fixed supply of bandwidth, they can only lease so much. We have agreed that the reservation amount per bandwidth leased is not relevant. The model you want to look at has a vertical supply curve. Finally, the theory that says consumer price should be the producer's marginal cost is predicated on the assumption that barriers to entry are low. This is not at all the case in ISPs. There is lots of expensive hardware, and there is a significant amount of technical expertise required. Thus, we cannot apply the standard econ 101 model, because we have violated its assumptions.

greggyb 8y ago

When an ISP leases bandwidth to a customer, they must reserve some amount out of their total available. They may reserve the full amount they have leased, or they may reserve some portion of it. Either way, the reserve is finite and proportional to the actual total bandwidth (the proportion may be 1:1, which is the simplest, but the proportion does not matter for this discussion). The marginal cost of provisioning me 100Mbps is therefore not limited to the cost of the hardware and setup, but is the opportunity cost of reserving bandwidth for my use. They may not lease this reserve to anyone else. This cost is reflected in the price. Additionally, we can observe that ISPs tend to grow. Even in a fixed area, they will upgrade hardware over time to increase total available bandwidth. This is driven by demand by existing customers for more bandwidth and new customers. This hardware is very expensive to provision. In fact, I expect that going from total bandwidth x to 2x likely costs more than twice as much as x. I would also argue that it likely costs at least thrice as much. Products of high fixed cost and finite supply will never be sold at marginal cost (or at least not for anything approaching profit). They will approach marginal cost as N increases (where N is the size of the finite supply). Another way we can think of a customer leasing bandwidth is that they are purchasing a share of that large and expensive network equipment. I posit that these are valid ways to look at it. The first, because there is not infinite bandwidth, ISPs must price it like the scarce good it is - prices are how we address scarcity. The second, because ISPs must not price their service as if they are never making another network upgrade. Again, I am willing to haggle on the cost of 2x performance. I would make all of these arguments for any value under 3x price, as I mentioned above, which is clearly stated in the original parent. There is no need to stick to a strict linear interpretation, and I have said as much more than once now.

sethrin 8y ago
For residential service, there is no reason to expect a 1:1 reservation. This would be extremely wasteful. The exact number is not relevant, but residential bandwidth is oversold as a matter of course. Talking about capex and aggregate costs also seems to be drifting from the main theme here: whatever it costs to double the network's capacity is going to be pretty circumstantial, and not directly relevant to the topic of consumer pricing. <i>Ceteris paribus</i>, the marginal cost per individual is still negligible, and the price per unit of provisioned bandwidth only loosely coupled to cost. ISPs charge extra for higher-tier packages because having only one tier of service is leaving money on the table, and because they can.
- greggyb 8y ago
  You're right. The reservation ratio doesn't matter - we agree on this and have both stated it's irrelevant. We're talking about average internet performance in a nation. This entire thread was kicked off by a comment about Akamai indicating the US as a whole has faster broadband than e.g. Germany. The cost to wire up the US with an average speed of 2x that of Germany is absolutely an influencing factor in the cost of bandwidth to a consumer. We are talking about performance, not widgets. To double the performance of a nation's network from p to 2p, it is not unreasonable to expect that the cost is greater than double or greater than 3x the cost of performance p. But, if you want to proceed from an analysis of marginal cost, we can do that as well. ISPs do not fit the model you are using. Marginal cost for something like an ISP is not a single number. There may be 10,000 consumers with a marginal cost near 0. Then customer 10,001 may have a marginal cost that is "install a new backhaul to the network backbone and extra 10G or 100G routing equipment to serve your neighborhood". Obviously customer 10,001 doesn't get stuck with a bill of $100Ks to be added to the network. You are attempting to pigeonhole an ISP into a model that assumes a single, flat fixed cost and (effectively) unlimited production of a good at a low marginal cost. That is not the industry we are in. The fixed costs of an ISP are a stepwise function, not a constant. You cannot model an ISP effectively with an assumption of constant fixed costs. All other things being equal, the marginal cost of a given customer is <cost of new network equipment or major network upgrade> / N, where N is the number of customers that can be served before the ISP has to install new routing hardware, or in the unhappy path lay new wire. Now, it is obvious that I have simplified things above. ISPs do not build infra for N customers, then wait for customer N+1 to build out the infra for the next N customers. ISPs invest in their infra in long planning cycles. They cannot effectively build out in this manner in response to a single customer. Thus, for years at a time, the supply of bandwidth is effectively fixed. With a fixed supply of bandwidth, they can only lease so much. We have agreed that the reservation amount per bandwidth leased is not relevant. The model you want to look at has a vertical supply curve. Finally, the theory that says consumer price should be the producer's marginal cost is predicated on the assumption that barriers to entry are low. This is not at all the case in ISPs. There is lots of expensive hardware, and there is a significant amount of technical expertise required. Thus, we cannot apply the standard econ 101 model, because we have violated its assumptions.