Hacker News Reader

Comment by hueving | Hacker News Reader

So is bandwidth though. I'm not sure where the idea comes from that ISPs have unlimited bandwidth.

Except if bandwidth were legitimately scarce ISPs would be trying to incentivize the use of bandwidth at low times and disincentivize its use when congested. Power companies can and do do this - the price per kwh can be much cheaper at night or in the morning than by day or evening when there is less peak load on the system.

Nowhere are IPSs offering "3mbps from 5-10 and 30mbps the rest of the day" plans. Because the cost to improve total network bandwidth is almost nothing, relative to the last mile costs. Every ISP could easily meet the full rated line bandwidth of every residential connection they have within their current margins without issue. They don't, because they have no competitive pressure to actually do anything than make sure the current service works. They have no competitors besides maybe cellular providers, who are already paying them in business contracts to attach towers to the backbone.

Our (rural wireless ISP) does something along these lines ... data between something like 1am and 7am does not count towards the monthly data cap.

At (peak) times they actively traffic shape and penalize the heaviest users. It sucks, but I don't see people willing to pay the $$$ to upgrade infrastructure faster.

> Nowhere are IPSs offering "3mbps from 5-10 and 30mbps the rest of the day" plans.

I've seen "off peak" data plans all over the world.

e.g. https://www.skymesh.net.au/nbn-services/fibre-to-the-premise...

    $54.95
    30GB Anytime Data
    60GB Off Peak
    50/20 Mbps*
    $1.83 per GB

A decade ago, my (European) ISP had even more complex rules: it was unmettered during the night, but during the day it had different caps for International, National and In-ISP traffic.

Aggregate bandwidth or rate is limited and is what costs money to increase but that's not what ISPs are charging for.

Do you think that everyone gets the aggregate rate they pay for? Do you think any ISP has capacity to deliver everyone their aggregate rate simultaneously? Does the combined bandwidth consumption of all customers over a given billing cycle approach the maximum theoretical bandwidth of the ISP? No, no, and no.

ISPs are charging for bandwidth like it's a commodity you buy at the store but if you buy a bag of rice at the store, the store doesn't take your left over rice at the end of the month and expect you to buy more. ISPs do though...

Paying for transfer isn't the same as paying for bandwidth. If I download something in the middle of the night vs at peak usage, it's the same transfer billing, but one utilizes bandwidth that isn't scarce and one does.

ISPs have profited massively from many many Moore iterations while consumer bandwidth is stuck on below 1990 technology.

Make no mistake, the technical side of this has only gotten cheaper. That's before you consider the "investment" necessary to resolve the vast majority of Netflix bandwidth issues is connecting a cable from one router to another in a CIX.

Of course when you do that it becomes blatantly obvious that ISPs have oversold and overbooked capacity dramatically as we get closer and closer to the last mile to the customer. Now they don't want to pay up to actually provide what they have sold.

> ISPs have profited massively from many many Moore iterations while consumer bandwidth is stuck on below 1990 technology.

On one hand they don't have to compete in many place so ISPs are close to being a monopoly so they can keep charging $80 for < 1Mpbs if they want.

At the same time Moore iterations don't dig trenches and lay fiber between cities. Also frequency spectrum doesn't grow with Moore's law either.And in many instances it is still mostly a shared resource. Cable companies usually have groups of houses share the upstream bandwidth.

Plain old copper has seen similar "appreciation":

https://www.sec.gov/Archives/edgar/data/1219210/000119312513...

Of course none of that ever makes it to the consumer, investment is stagnant or down and all this tech ends up achieving is accommodating higher Internet penetration (and customer counts).

Here is the DOCSIS evolution:

http://www.cablelabs.com/full-duplex-docsis/

All this cable infrastructure previously wasted on free to air crap has turned into a veritable gold mine with no further investment and tech again making up for all the growth so you don't have to spend any.

Sure, Moore isn't helping dig trenches but he sure has been tremendously helpful in turning that ole shitty copper wiring financed by Uncle Sam into a lot of green.

This is completely ignorant on two fronts. First, the neither POTS nor cable was "financed by Uncle Sam." Second, new headend equipment isn't free, and DOCSIS upgrades have gone hand-in-hand with node splits and pushing fiber deeper into the HFC network.

> ISPs have profited massively from many many Moore iterations while consumer bandwidth is stuck on below 1990 technology.

On what planet? I did a speedtest.net run the other day on LTE and got 150 Mbps down/20 Mbps up near a strip mall in exurban Maryland. That network did not come cheap.

The effective bandwidth of that network even on some unobtanium 50 GB plan is <1 Mbps and that's before we get into latency and packet loss discussions.

Nice distraction, but no cigar.

That's like saying the "effective speed" of a 777 is 45 mph because that's the maximum average speed over a month accounting for refueling and servicing. Or that the "effective battery life" of a MacBook Pro is 70 minutes because that's how long you can run it at full tilt. It completely ignores the typical customer use case.

I feel nerdsniped! The utilization rate of a major airline 777 is ~12 hours per day. If the ORD-LAS flight I took today is representative, average in-transit speed is 331mph, so the average speed of a jetliner isn't 45mph, but rather... 165mph?

I'm embarrassed to admit I didn't even try to do the math.

Hmm, I think that figure could be off by a decent amount though depending on what exactly is measured by "utilization". Taxiing, for example, or sitting at the gate with an hour delay.

I think the best way to figure this out is to conduct a little study: set up a script to monitor a representative sample of planes on FlightAware for a while.

Taxiing is included in utilization time, but also in my average transit time (I just took the nautical distance traveled and divided it by the actual transit time).

Moore's law has not held for long-haul bandwidth for quite some time. Moore's law is for semiconductors and that has little to do with signal efficiency.

Reply on news.ycombinator.com