Feeling ripped off by your ISP? It's getting cheaper to pipe your packets globally The cost of IP transiting – the cost to transfer internet data across ISPs and other large public networks – continues to fall. This mean it's getting cheaper to send webpages, messages and other traffic. London has seen a 22 per cent price drop to just $1 per Mbps per month. As a global average, in the past three years, the …
Obama's pivot to Asia is as much about commercial (political economic) center of gravity as militarily. I would expect much more internet investment to follow. One fly in that ointment though is the PRC government's control issues. Certainly (I believe) the margins are there so long as we don't have a capacity over-investment as happened in the '90's. BTW, whatever happened to all that "dark-fibre"? Not an industry I usually track. My bad!
Most of the Dark Fiber left only works at the 100 Mb/s level or lower. Long distance cables are a lot more then a piece of glass, there are repeaters every 100-200 Km, and those are the bottleneck. Since it takes about the same amount of energy to power a 100 Mb link as a 10 Gb link, companies are just laying new fiber that can accommodate 1,000-10,000x the bandwidth with the same operating cost, which is why NSPs are making mountains of cash while their fees keep falling. It works out that if they replace a piece of fiber with 10x the capacity, they can easily charge half as much but make 5x the profit (Assuming 100% usage) and even subtracting the payments on the loan to lay the cable, they still come out on top with a nice profit increase.
As for those old fibers, a lot of them have failed over time, and the rest get leased to companies to run private site-to-site links (mostly telcos for their long-distance back-hauls given the guaranteed bandwidth).
I mean optics and routers have a limited lifespan. After 10-20 years maintenance costs will rise, which means you will want to replace them. Also if you want to get rid of your equipment earlier, there's a decent refurbishing market so you can recover a large fraction of your costs.
Plus equipment costs pale in comparison to the wages of the people maintaining them. A network engineer will easily cost you 100k$ a year if you count all costs directly associated with hiring a person.
It's not a unit per se, you pay for capacity not usage. So 1Mbps per second of capacity for a whole month is $1.
If you want to work back from to get the per MB the value would be 1 ($) divided by the total number of bytes that _could_ be delivered which is 1 (Mbps) divided 8 (bits to get number of bytes you can transfer in one second) multiplied by 2,592,000 (assuming a 30 day month).
1 / ( (1 / 8) * 2,592,000) = 1 / 324,000 = $0.000003086 or $3.086x10-6 per Megabyte
which could be expressed as $3.24 per TB
So it really does cost over $2 million pa for a 10GigE port, then? (Disclaimer: I've no idea what the actual cost is, but it sounds high to me, in my ignorance.) And if it's based only on nominal capacity rather than usage, why not simply quote the price for the port rather than convert it to this odd unit?
[Paris: because she probably understands this better than I do.]
Excluding bulk discounts, that would be about right. Consider this scenario and it will make a lot of sense:
--In a metropolitan area an ISP may have 1 million customers, they provide each customer with a 20 Mbps link and charge $25/month for that link. This would produce $300 million dollars per year in income for a theoretical 240 Gb/s (valued at $480 Million of bandwidth) meaning a loss of $180m in just transit cost
That is the theoretical, however very, very few people actually use up all the bandwidth they pay for, so:
--Based on the fact that most ISPs have a 500 GB/month data cap, which works out to just under 0.2 Mb/s per connection. So now to feed that much bandwidth to would only take 2.4 Gb/s link for transit, so 2x 10 Gb/s links would be more than sufficient to handle spikes and heavy users. So this theoretical ISP would be receiving $296m in income after transit fees are subtracted.
Now consider that the ISP is also probably doing quite a bit of packet optimization and that a lot of packets don't actually leave their own network given their peering agreements with the big bandwidth consumers, and you see that $2m per year for a 10 Gb/s connection is nothing for them.
In the networking world we always talk about bits, not bytes.
It makes no sense to talk about ports because your theoretical 10G port will almost certainly be shared. You might have a 100M or a 1G VLAN on that port but not the whole thing. Depending on who you peer with or how you've arranged for traffic to be managed you might flex that connection up and down in size - sometimes even dynamically. You pay for capacity but no-one likes to pay for unused capacity and so a lot of effort goes into making the paid-for capacity as close to the actual usage as possible.
In a market where ISPs are selected by customers based purely on lowest cost, anticipating demand and adding bandwidth ahead of the demand curve is a route to bankruptcy.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
