Hacker News Reader

Comment by endisneigh | Hacker News Reader

Why shouldn’t there be a cap? Bandwidth isn’t free.

It’s sad seeing people on a forum called hacker news having no idea how networking infrastructure works.

Because it's next to free (no additional cost to run the infrastructure we're already running) and we're nowhere near the limits of our existing infrastructure's overall bandwidth limits right now, as evidenced by the lack of caps not affecting quality of internet service for the entirety of 2020, when more people were online per hour than at any previous point in history.

Source? Running things requires labor, labor is getting expensive. Pricing isn’t set on simply the marginal cost of materials…

Running a fully utilized fiber line is.. exactly the same amount of labor as running a 10% utilized one.

The cost to track and bill people for caps and let them pay overages though..

Chokepoints are peering exchanges, transit handoffs, and core network gear, which is sized for a percentile of utilization below constant 100%. Having implemented a billing system and utilization monitoring for managed hosting datacenter environment, it is trivial to collect metrics from equipment and bill accordingly. You don't want to nickle and dime or gouge customers, but you do want to ensure proportionality across the customer population.

Some combination of caps, overage charges, traffic shaping, and edge CDN appliances is needed (Netflix, Akamai, etc). Off peak unlimited is also a potential strategy to drive low priority transfer to low utilization periods. Pick your poison. There is a reason most of Netflix global infra is monster CDN boxes closest to customers.

https://openconnect.netflix.com/en/

> There is a reason most of Netflix global infra is monster CDN boxes closest to customers.

Because it would be ridiculously expensive to egress video streams directly from AWS. They've determined that it's cheaper to pay ISPs for rack space.

You misunderstand the model. They don't pay ISPs, they offer the appliance for free to the ISP for colocation, reducing the upstream networking costs. AWS egress being extortionate is a distinct issue. The cost to the ISP in the rack (cooling, power, space) is immaterial considering the transfer savings (not only opex, but reduced capex from needing to oversize network gear or specific ports for peak streaming consumption) over the life of the relationship.

If you want to learn more, lurk on the NANOG (North American Network Operators' Group) mailing list. It is common to monitor AS traffic flows to understand whether you can offload with CDN appliance relationships or how to improve your network architecture (which networks to peer to directly or which internet exchanges [IXs] to meetup at). These agreements sometimes happen at a bar over beers during NANOG events. Call Verizon and ask them how much a 1Gbps circuit is. Consider why Comcast built their own national backbone.

Also, some important history to remember. Fast.com measures your speeds to Netflix servers because some ISPs wanted to charge their customers and Netflix for that transfer (or to upgrade peering points) because of the revenue destruction from cable customers fleeing for streaming (net neutrality debate).

https://archive.nanog.org/meetings/nanog45/presentations/Nor... (Peering 101, Bill Norton [Equinix cofounder], NANOG 45)

https://drpeering.net/white-papers/Video-Internet-The-Next-W... (Video Internet: The Next Wave of Massive Disruption to the U.S. Peering Ecosystem (v1.7))

https://drpeering.net/HTML_IPP/ipptoc.html ("The Internet Peering Playbook")

https://arstechnica.com/features/2008/09/peering-and-transit... ("How the ‘Net works: an introduction to peering and transit")

https://arstechnica.com/information-technology/2014/07/how-c... ("How Comcast became a powerful—and controversial—part of the Internet backbone")

https://openconnect.zendesk.com/hc/en-us/articles/3600361636...

> How much does the appliance cost my organization?

> Appliances, including replacement appliances, are provided to qualifying ISPs at no charge when used within the terms of the license agreements.

Also, when caps are implemented, they're ridiculously expensive and limited, especially in an age of streaming video. Expensive comically beyond any reasonable conception of the actual marginal cost of transiting the traffic.

Also - just look at LTE operators. They're also no-cap at this point (though subject to QoS at certain breakpoints, but they're typically reasonable breakpoints ime?). And that's in a SIGNIFICANTLY more capital intensive market - you have most of the concerns with terrestrial fiber, plus the joy of having to own RF spectrum and maintain towers etc.

So I think no-caps-and-required-speed requirements would just make sense and be required to keep ISPs from trying to backtrack on being forcibly dragged into the modern area like balky calves.

I think that the issue is lack of competition. on mobile market it's easy to switch operator to any from multiple nation wide ones. for cable ISP AFAIK in US there is usually one. (in my central european country there is at least 3 ISP available ans AFAIK fibers to building are in some way rented to internet providers (?))

In principle caps help with oversubscribing. If you want to use a 10GBit/s uplink to serve 5000 people with 100MBit/s each, you won't saturate your uplink nearly as often if you add a data cap.

Of course that's not all that different from restricting bandwidth, just that you are restricting average bandwidth while allowing for some burst. Still violates the spirit of having fast bandwidth, unless the caps are quite generous.

They do help push people to use less, but for places like Comcast I see data caps as just a revenue stream increaser. The entire company strategy seems to follow this approach.

Yes and no, depending on your plant build out you might have great interplant connectivity - 10G between each switch, 10G to the router, but the extraplant connections still cost and have bandwidth limits. So if you only use 2 gbps total during peak, you may not need to upgrade your plant to support higher speeds but you will have to pay more to your upstream provider to support it.

This is actually relevant to my ISP day job, we have a pipe to a well known backbone provider, and dual pipes to a statewide provider. We ran on the dual pipes for a long time before pushing our ISP traffic over the single backbone provider. These days, if that main ISP pipe goes down, it fails over to the dual pipes to keep everything running. But now that we are offering packages higher than 25/3 for DSL, terrestrial wireless, and cable, the dual pipes don't have enough bandwidth to keep up.

If we offer gigabit service over fiber (and we do) just a couple of customers actively using their whole pipe is an enormous chunk of the network compared to the 40 or so customers that same gigabit of bandwidth would serve on previous 25/3 packages. We don't have data caps or even contracts (small town benefits :P ) but there is a lot more to it for smaller ISPs than adjusting the rate limiting and packages we allow people to use.

It’s true that smaller ISPs have their challenges, but the cost for both interplant and extraplant connections has been generally decreasing. Economies of scale and peering agreements can mitigate many of these costs.

True, we connect our main ISP pipe to an internet exchange for that reason. It's also interesting to see how the industry is adopting pluggables for a lot of newer deployments because depending on what you need, there's a lot of flexibility now compared to even 5 years ago.

> Because it's next to free (no additional cost to run the infrastructure we're already running)

Aren't we effectively arguing CapEx versus OpEx?

Sure, an ISP's OpEx is tiny. But CapEx is huge, especially when you've got an old neighborhood that's all wired up with copper and you gotta replace it all with fiber going into the house.

Cost of the traffic is: hardware (staggered cost covered over many months), and running costs (subscription to higher tier internet carriers, electricity, rent of the premises, salaries etc). Most of this is covered by the monthly fees, to have a profitable business one must have those costs covered as a minimum. None of the above affects how much internet data goes through, zero or maxed out capacity, cost is the same.

This is patently false. Given optics can only support so many people at a given latency and bandwidth.

Let me throw that question back at you: why should there be a cap? You are paying ISPs for a certain bandwidth speed per month and they sure as hell are not prorating your service based on data that you did not consume. If the ISP is in capable of delivering the agreed-upon band with every minute of the whole month, then they are not meeting the SLA that you’re paying them for.

If ISPs were like an electric utility, we'd see something like a cost per GB transferred in cents, with a minimum charge reflecting the cost of maintaining a connection of a certain speed to the network.

Which is algebraically identical to a monthly charge and data cap with overage charge. The main issue is the overage charge is too high, it should be like 1 cent per GB (Comcast is charging 20x that).

1c/GB would be a decent rate for home data transfer in 2023, but where would you set the free limit? For reference, at my house I have a 1Gbps line and it looks like in 48 hours I have downloaded 1TB of data (looking at the number of received bytes my AT&T device is reporting). Am I normal? I don't know, but AT&T does...

I think that's rather high compared to average, considering the standard cap for Comcast Xfinity residential is 1.2TiB per month, and they claim only "a very small percentage" of their customers use more.

I don't think the actual cap really matters if the per-GB and base pricing reflects the true costs. If it's low it means heavy users pay more, if it's high, light users pay more.

1TB in 48 hours is absolutely not a normal residential connection usage.

The average US residential customer uses a bit over 500GB/mo in data. You're doing 2x that in 1/15th the time.

https://www.allconnect.com/blog/report-internet-use-over-hal...

They already charge for bandwidth. People pay more for 500mpbs than 50mbps. We are talking about total monthly data caps, which have nothing to do with bandwidth.

False. A month is a discrete amount of time that yields a potential maximum amount of data downloaded, this amount is clearly related to the bandwidth provided, hence caps.

To think otherwise is to falsely believe they support unlimited throughput. Furthermore the bandwidth itself literally is a function of the transmission material and infrastructure, which also isn’t free…

I have a 500 Mbps connection with a 1TB data cap, both pretty standard numbers in the US. If I used the connection at its advertised speeds, I would exhaust my monthly limit in about 4 hours. So no, that relationship is non-existent. The data cap math instead is "99% of our users use < 1TB, so we will charge the rest more because we can".

Or more accurately, to disincentive excessive use. They don’t need a cap to simply charge you more money, they’ve been doing that just fine with or without caps.

That would be a valid argument if they defined "excessive" based on infrastructure limits rather than a demand curve and market research.

All prices are based on “demand curve and market research” so I’m not sure what your point is.

My point from the start has been that ISPs put data caps and overage charges in place to make extra money, not to protect their infrastructure.

If that was a big issue, why are smaller ISPs (like sonic.com) able to offer 10Gb ethernet at a flat rate with no caps?

Peering agreements at internet exchanges make additional bandwidth very cheap to provide and infra costs are spread over years and many users, further reducing ongoing costs..

Bandwidth caps not only have a weak fiscal basis but also clash with net neutrality tenets, go against what made the internet successful in the first place.

Bits that go unused have no value. Caps on numbers of bits in a billing cycle don't make sense. While it isn't exactly pure artificial scarcity, it's darn close. A retail ISP already has to have some fair share mechanisms for peak use times. That leaves very very little usefulness in monthly caps.

once the infrastructure is there does each byte/gigabyte really have an intrinsic cost? isn't more about the bytes per second and not the bytes total?

There’s more to running an isp than simply plumbing electricity.

sure, but is your 'profit' to be made in.. charging per byte consumed/sent? it just feels... gougey?

I had to live under an ISP that charged something on the order of $10+/GB after a certain amount and after they were forced to offer unlimited internet it was an actual breath of fresh air just to be able to use the internet and not worry about literally everything in the house, or if I could update my OS, or play this game, or watch this Netflix show, etc.

i'm honestly curious. why not a standard margin on your 'per second' allowances (100/100Mbps, etc) or other 'addon' services?

why should someone with a ring doorbell (or a homekit secure video whatever) be punished over someone who doesn't?

Sure, but the argument for caps is that it reduces overall demand and therefore peak consumption, and peak consumption is what the ISPs actually have to pay for.

hmm, spitballing here but if they were really suffering could they not offer a package with lower speeds during peak? (while slowly using some of their profit to re-invest in infrastructure?)

"It's weird that none of the so-called experts agree with me. The problem is clearly with their understanding of the space, not mine."

Reply on news.ycombinator.com