Who's paying for this?
October 18, 2012 1:29 PM Subscribe
What percentage of my monthly ISP bill pays for the internet infrastructure? What percent of the internet is paid for by home users and business users as consumers, versus business users who are serving content, like Facebook or NetFlix Streaming? I would love a pie chart, but whatever you got is appreciated...
It's more of a pyramid, but not a pyramid scheme-- you only pay money to the next level, not to the top. You pay your ISP for access to the internet at large, as well as service for a line to your home or business. They, in turn, pay one or more "middle tier" companies who own the high-speed lines that carry data across parts of the city. To connect between cities, they pay tier 2 and tier 1 companies that operate the inter-city, interstate "internet backbone" lines, which are the major trunks, outrageously wide and fast connections that carry data around the country. Likewise, there are privately owned international cables, including plenty of undersea cables, that do the same.
Your ISP and everyone above them pays for A) usage and B) premium positioning. By the positioning, I mean, are your servers located in a data center with 10 T1 lines (15 megabits) for the whole buiding, or are they setting on the head-end (arrival point) of the 100 Gigabit undersea cable that connects Florida to Spain?
Your ISP has to pay someone for yours and everyone else's usage, but they charge you a consumer market rate based on their estimate of your usage (or the usage they keep you to by throttling your connection), plus you do keep on the lights and fund the payroll of your ISP, which provides support, etc.
Google, Facebook and Netflix deal directly with the upper tier owners, and while everyone does benefit from having those 3 around in the general standard-of-living sense, each of those companies has a screamingly huge internet bill because they pay for positioning and bandwidth at rates that stagger the mind. Microsoft's bills from their weekly Tuesday patch releases must also be some whoppers, because they have to scale up their usage immensely for 1 day of the week.
posted by Sunburnt at 2:50 PM on October 18, 2012
Your ISP and everyone above them pays for A) usage and B) premium positioning. By the positioning, I mean, are your servers located in a data center with 10 T1 lines (15 megabits) for the whole buiding, or are they setting on the head-end (arrival point) of the 100 Gigabit undersea cable that connects Florida to Spain?
Your ISP has to pay someone for yours and everyone else's usage, but they charge you a consumer market rate based on their estimate of your usage (or the usage they keep you to by throttling your connection), plus you do keep on the lights and fund the payroll of your ISP, which provides support, etc.
Google, Facebook and Netflix deal directly with the upper tier owners, and while everyone does benefit from having those 3 around in the general standard-of-living sense, each of those companies has a screamingly huge internet bill because they pay for positioning and bandwidth at rates that stagger the mind. Microsoft's bills from their weekly Tuesday patch releases must also be some whoppers, because they have to scale up their usage immensely for 1 day of the week.
posted by Sunburnt at 2:50 PM on October 18, 2012
I'll just add that most large ISPs peer with large content providers, such as YouTube and Facebook, and as far as I know, no money changes hands for that.
posted by empath at 2:56 PM on October 18, 2012 [1 favorite]
posted by empath at 2:56 PM on October 18, 2012 [1 favorite]
I work for a small ISP and we struggle to answer these types of questions just for purpose of modelling our own costs (to set prices, to trade off install charges vs contract terms, etc). So to expand the question outside of one small ISP and apply it to "the Internet" as a whole is daunting.
Some realities I can attest to:
-Wholesale pricing for "Internet" (known as "transit") ranges from a couple $ per Mbit/s down to well under $1/Mbit/s when delivered at a major carrier hotel in the US. As you purchase gbit/s up to 10s or 100s of gbit/s the cost goes down a lot. But keep in mind that a large network that would tend to buy those large volumes is also going to have an easier time negotiating settlement-free interconnection (peering, also known as SFI).
-Carriers, content providers, and eyeball networks are constantly jockeying prices and who they'll interconnect with for free, and the geography of those interconnections. For example, if you've watched comapnies like facebook or google grow over the past decade you can see how they balance building out their own backbone vs purchasing transit from established carriers. It's a business trade-off. Google pretty much is its own global backbone at this point. Twitter, when small, purchased most transit from NTT, but is branching out more now.
-You'll then pay $100-$300/month for a fiber optic crossconnect good for 1 to 10's of gbits/s within the carrier hotel. Notice at low levels (<1Gbit/s) this makes interconnecting between networks at a carrier hotel cost prohibitive, even if you disregard the cost of the equipment (routers/switches).
-You'll then pay a hugely variable cost to transport those bits to some other city where consumer and business customers are. In well-built-out areas (say, parts of California) in small volumes this can be in the $4-figures per 10gbit/s per month to go between metro areas for 100-300 miles. Within a metro area (saw, SF bay area) the cost is a *lot* less due to tons of competition. In other less-built-out areas you end up having to build yourself or pay a carrier to build and light fiber to get to cities or parts of cities, and you merge the construction cost into 3 or 5-yr contracts (for short distances) or 10-20-yr contracts for entire strands of fiber (for 100+ mile long-haul).
Getting down to a more micro-level you have copper plants and/or metro-fiber routes in individual cities/towns, and costs vary hugely there depending on the amount of competition. Add in statistical multiplexing (time-based over-subscribing of an ISPs network connections) and different patterns between, say, colocated servers pushing out content 24x7 versus a consumer DSL line where it's idle all day and then the user gets home at 6pm and starts pulling netflix and facebook content. At that micro level you have other costs too -- leased copper for the DSL, leased building for the colocation customer, power, air conditioning, maybe even metro fiber (leased or constructed then owned over 20 years) to a street-side cabinet that then muxes out to more metro-fiber or DSL customers.
So as you can see, modeling it all ends up being like predicting the weather.
posted by frontmn23 at 3:31 PM on October 18, 2012 [4 favorites]
Some realities I can attest to:
-Wholesale pricing for "Internet" (known as "transit") ranges from a couple $ per Mbit/s down to well under $1/Mbit/s when delivered at a major carrier hotel in the US. As you purchase gbit/s up to 10s or 100s of gbit/s the cost goes down a lot. But keep in mind that a large network that would tend to buy those large volumes is also going to have an easier time negotiating settlement-free interconnection (peering, also known as SFI).
-Carriers, content providers, and eyeball networks are constantly jockeying prices and who they'll interconnect with for free, and the geography of those interconnections. For example, if you've watched comapnies like facebook or google grow over the past decade you can see how they balance building out their own backbone vs purchasing transit from established carriers. It's a business trade-off. Google pretty much is its own global backbone at this point. Twitter, when small, purchased most transit from NTT, but is branching out more now.
-You'll then pay $100-$300/month for a fiber optic crossconnect good for 1 to 10's of gbits/s within the carrier hotel. Notice at low levels (<1Gbit/s) this makes interconnecting between networks at a carrier hotel cost prohibitive, even if you disregard the cost of the equipment (routers/switches).
-You'll then pay a hugely variable cost to transport those bits to some other city where consumer and business customers are. In well-built-out areas (say, parts of California) in small volumes this can be in the $4-figures per 10gbit/s per month to go between metro areas for 100-300 miles. Within a metro area (saw, SF bay area) the cost is a *lot* less due to tons of competition. In other less-built-out areas you end up having to build yourself or pay a carrier to build and light fiber to get to cities or parts of cities, and you merge the construction cost into 3 or 5-yr contracts (for short distances) or 10-20-yr contracts for entire strands of fiber (for 100+ mile long-haul).
Getting down to a more micro-level you have copper plants and/or metro-fiber routes in individual cities/towns, and costs vary hugely there depending on the amount of competition. Add in statistical multiplexing (time-based over-subscribing of an ISPs network connections) and different patterns between, say, colocated servers pushing out content 24x7 versus a consumer DSL line where it's idle all day and then the user gets home at 6pm and starts pulling netflix and facebook content. At that micro level you have other costs too -- leased copper for the DSL, leased building for the colocation customer, power, air conditioning, maybe even metro fiber (leased or constructed then owned over 20 years) to a street-side cabinet that then muxes out to more metro-fiber or DSL customers.
So as you can see, modeling it all ends up being like predicting the weather.
posted by frontmn23 at 3:31 PM on October 18, 2012 [4 favorites]
I've worked for and with companies that build and run Internet infrastructure such as carrier hotels, subsea cables and national fiber networks. Fronton23's answer is generally excellent as to why there's no easy answer to this question.
To his answer, I would add that the thing to understand is that "the Internet" makes up only a portion of the data travelling across these networks and stored in these data centers. Much more of it is private networks of some sort run by businesses, governments and others large entities who want to transport data much more securely and reliably than happens on the Internet.
Some of this connectivity can be charged at pretty astronomical prices - if for instance, you own the lowest-latency route between New York and London and are selling to high-frequency traders, or if you're selling bandwidth to/from some place like Southern Africa where there may be only one or two cables that land in that region. In other cases, say, for instance, most Trans-Atlantic routes between Europe and the US, pricing is quite low due to high levels of competition.
But generally, more money comes from large companies of all sorts who are willing to pay a premium for guaranteed performance, availability and security, versus the Internet which is what is known as "best effort".
So if you're making a business case for laying a new cable between New York and Brazil, you're going to start by looking at whether big companies will pay a premium for a faster connection or more bandwidth between those two locations. You'll make a lot of money on that traffic, at least for a few years until someone else puts in more capacity and price wars drive down your margins. You'll make much less money on Internet traffic, but you can't fill the entire cable with premium traffic so you count on Internet traffic to bring in additional volume that lowers your cost per unit.
The Cisco Visual Networking Index doesn't get at the money aspect, but is a good place to go to understand what types of traffic from what sources are moving across the Internet, and even includes some charts and graphs.
posted by psycheslamp at 5:23 PM on October 18, 2012
To his answer, I would add that the thing to understand is that "the Internet" makes up only a portion of the data travelling across these networks and stored in these data centers. Much more of it is private networks of some sort run by businesses, governments and others large entities who want to transport data much more securely and reliably than happens on the Internet.
Some of this connectivity can be charged at pretty astronomical prices - if for instance, you own the lowest-latency route between New York and London and are selling to high-frequency traders, or if you're selling bandwidth to/from some place like Southern Africa where there may be only one or two cables that land in that region. In other cases, say, for instance, most Trans-Atlantic routes between Europe and the US, pricing is quite low due to high levels of competition.
But generally, more money comes from large companies of all sorts who are willing to pay a premium for guaranteed performance, availability and security, versus the Internet which is what is known as "best effort".
So if you're making a business case for laying a new cable between New York and Brazil, you're going to start by looking at whether big companies will pay a premium for a faster connection or more bandwidth between those two locations. You'll make a lot of money on that traffic, at least for a few years until someone else puts in more capacity and price wars drive down your margins. You'll make much less money on Internet traffic, but you can't fill the entire cable with premium traffic so you count on Internet traffic to bring in additional volume that lowers your cost per unit.
The Cisco Visual Networking Index doesn't get at the money aspect, but is a good place to go to understand what types of traffic from what sources are moving across the Internet, and even includes some charts and graphs.
posted by psycheslamp at 5:23 PM on October 18, 2012
Well, here are some broad strokes about what it costs to move data around the Internet. Akamai Technologies charges (I've read, can't find the cite) $.12/GB (gigabyte) to serve content. You upload a file to them, they let your customers download it. That's the hop from Akamai's servers (which are deliberately close to the "edge" of the network, to avoid going over the public backbone, which costs and is slower) from the customer's ISP. Amazon's web services (S3) charges the same for bandwidth up to 10TB (terabytes = 1000 GB) a month, down to $0.05/GB for 500TB/month, so it does get cheaper if you're a large ISP.
Data caps in the U.S. start at around 250GB of data, so let's assume that your ISP is paying Amazon's best rate or better, so a big 99th percentile downloader would incur costs of 250*.05 = $12.50 a month. Most customers use far, far less than their cap, so a reasonable average cost for external bandwidth would be $5 or so a month. There are definitely costs for going the "last mile" from the ISP to your house, but given that the data can go halfway around the world for a nickel, you're talking $10 a month for our hypothetical average customer. ISPs in the U.S. are making out like bandits, basically, because there's so little competition. Look overseas where countries like France offer 28 Mbs connections for $45 a month.
posted by wnissen at 8:36 AM on October 21, 2012
Data caps in the U.S. start at around 250GB of data, so let's assume that your ISP is paying Amazon's best rate or better, so a big 99th percentile downloader would incur costs of 250*.05 = $12.50 a month. Most customers use far, far less than their cap, so a reasonable average cost for external bandwidth would be $5 or so a month. There are definitely costs for going the "last mile" from the ISP to your house, but given that the data can go halfway around the world for a nickel, you're talking $10 a month for our hypothetical average customer. ISPs in the U.S. are making out like bandits, basically, because there's so little competition. Look overseas where countries like France offer 28 Mbs connections for $45 a month.
posted by wnissen at 8:36 AM on October 21, 2012
« Older Sales as a bridge to other careers? | It's one of those things that you're always... Newer »
This thread is closed to new comments.
Other than that, I think you probably need to be more specific about what you're asking about. Every ISP is different, and clecs have different strucutures than ilecs or cable companies, and some ISPs focus on business customers and others focus on residential.
I think the largest cost by FAR for ISPs is going to be labor-- everything from sales to engineering to the field techs we send out on installs. At the dsl provider I worked for, we lost money in every single customer we signed up until they were with us for something like a year, just on the set up and equipment costs.
After that, the biggest cost is going to be running fiber and so on. The actual routers and switches, while expensive, are peanuts compared to the cost of running fiber through a city.
As far as business vs residential, I don't know the exact breakdown, but business customers as individuals generally pay much, much higher costs for the same bandwidth as residential customers, and ISPs spend a lot more money supporting them.
posted by empath at 2:37 PM on October 18, 2012