Can anyone explain to me what the marginal costs are for additional traffic, within Comcast's network?
I presume that once you're powering the routers, they consume pretty equivalent power at peak and off-peak times, but maybe that's a faulty assumption?
> Can anyone explain to me what the marginal costs are for additional traffic, within Comcast's network?
Marginal costs within the design capacity of the network are close to zero. If the traffic happens to be external, to a network Comcast pays transit to (if such exist) and it's in a direction that's billable (transit is paid in the direction that is larger) then the wholesale rate (for that fraction of the traffic) would be billed at about 0.1 cents per GB.
As I understand it the marginal cost of the first N gigabytes is approximately zilch, but N+1 gigabytes is $$$$- because they now have to upgrade the infrastructure.
The price tags they put on more data are almost certainly an attempt at "market allocation" rather than the actual cost of the data. In other words, the increased price on bandwidth lowers demand, hopefully to the level Comcast is targeting. Think of electricity companies, and off-peak rates. They charge more during peak load to drive the market to shift more load to off-peak times.
> As I understand it the marginal cost of the first N gigabytes is approximately zilch, but N+1 gigabytes is $$$$- because they now have to upgrade the infrastructure.
Missed out a step there. It's not zero then suddenly $$$$. It's more a question of gradual, usually low cost, equipment upgrades. All backhaul is via fiber optics, so by switching optics you go from 1G/10G/100G to multiple 1G/10G/100G waves. You might need a new router blade or a bigger chassis at some point.
Only if you really messed up do you need infrastructure upgrades. This might mean splitting nodes, i.e. pulling a bit more fiber to hotspots, but that is why you have spare empty conduits and excess fiber strands in your cables. If you don't then you truly deserve all the misery you brought upon yourself.
I would make the same assumption. The issue is they oversubscribe their capacity. And rather than upgrade their network to actually support what they sold, they want to introduce usage caps.
Everyone does. That's standard in the trade, because not everyone will be using their connection at the same time. There are huge savings in only buying enough bandwidth to handle, say, your 95% load and not enough to satisfy all of your customers at once.
My problem is that the ISP I work at had about a 10:1 oversubscription ratio. Comcast wants to fine individual customers who exceed 70:1, so they're probably aiming for a 200:1 overall ratio. That's just highway robbery.
> There are huge savings in only buying enough bandwidth to handle, say, your 95% load and not enough to satisfy all of your customers at once.
It's bad engineering to run your links hot at 95%. (Here's looking at you, Comcast)
You want to keep link usage at under 50% in order to have adequate margin for growth, be able to fall over to your secondary links without saturating them for maintenance/repairs/emergencies and handle microbursts which don't show up on the 5 minute average timescales monitoring is done at.
Part of the problem is adding capacity in an HFC network is expensive and difficult, eventually adding more carriers is not enough, and you needs to break the broadcast domain down even further, which means adding a new node - sometimes that can be just adding cards to a cabinet, others its a new physical box that must be permitted and placed - a process that can (and often does) take years.
xDSL has it easy, they can just turn up more backhaul at the CO, which is often a process that takes no more that days, if not hours.
> Part of the problem is adding capacity in an HFC network is expensive and difficult
Sure, but no more than doing VDSL, which is basically the same thing for ADSL as splitting nodes in a HFC network. A FTTC+VDSL upgrade is about $400 per subscriber which is $17 per month on a typical 24 month contract. That's quite doable and typical network life cycles are much longer, which brings the cost further down.
When you build out VDSL, you're expecting to provide service to every customer in range at a fixed oversell ratio, if you need more bandwidth you can turn up additional fibers to the node, at the worst, split the mux down to smaller numbers of subs per, but the aggregation point never changes - it's always at the RSLAM.
With cable, the point of aggregation is at the subscriber prem, just like thinnet - the only way to get more bandwidth to the subscriber is to decrease the size of the broadcast domain by adding another node.
It's basically my contention that like most wireless networks the pinch point is access layer, not in the backhaul, HFC networks are much like wireless networks, they have a fixed amount of RF bandwidth shared with N subscribers over a given area - being that the amount of spectrum a given area can use is fixed (by licensing or by bandwidth limitations in the cable) the only way to handle more subscribers in a given area is to decrease the number of subscribers per network element, in short, add more sites, or nodes.
I presume that once you're powering the routers, they consume pretty equivalent power at peak and off-peak times, but maybe that's a faulty assumption?