If you phrase it as just "don't prioritize traffic", you've just broken most CDNs, as well as QoS that optimizes for latency/bandwidth to improve VoIP and similar.
There's absolutely nothing wrong with, for instance, a content provider handing an ISP a box that sits on their network, mirrors/caches content, and serves it up to that ISP's customers. That helps the content provider, the ISP, and the ISP's customers. And I don't think the majority of "net neutrality" advocates want to break that.
When people talk about "net neutrality", the problems they have in mind tend to look like either an ISP extorting content providers to extract extra revenue, or an ISP giving their own first-party services an advantage by deliberately degrading third-party services. For instance, see Comcast and Netflix.
But even then, how do you draw a bright line between "deliberately degrading", "not giving an advantage to", and "not doing anything to address bandwidth problems when they arise"? (Comcast could claim the latter regarding Netflix, even though it seems obvious what it means when they don't arrange additional bandwidth on those saturated links.) What if the ISP provides the CDN infrastructure, rather than the content provider? What if the ISP contracts with a third-party CDN?
How would you suggest writing a regulation that doesn't break CDNs and similar, and that can tell the difference between "deliberately hurting" and "not helping"? How will that regulation adapt when people come up with new ideas to make the web faster?
(One angle I could imagine working: limiting the ability of one company or family of companies to run both an ISP and network services via that ISP. Still hard to write something robust, though.)
> There's absolutely nothing wrong with, for instance, a content provider handing an ISP a box that sits on their network, mirrors/caches content, and serves it up to that ISP's customers. That helps the content provider, the ISP, and the ISP's customers. And I don't think the majority of "net neutrality" advocates want to break that.
...until the ISP decides to charge a large amount of money for the ability to do that, and then under-provision their peering links so that there is no other way to have adequate performance.
But there is a simple way to fix that. Require last mile ISPs to do free peering with anyone who has traffic for their last mile customers. If you bring traffic directly to their local central office, they have to take it and deliver it up to the speed that they've sold to their customers.
Then they can offer all the CDN service they like, which people will use as long as they're charging market rates but decline and rely on peering (and install their CDN equipment across the street) if they try to charge monopoly prices.
> ...until the ISP decides to charge a large amount of money for the ability to do that, and then under-provision their peering links so that there is no other way to have adequate performance.
The latter is the primary problem here.
> Require last mile ISPs to do free peering with anyone who has traffic for their last mile customers.
What does "last mile ISP" mean here? DSL is no longer competitive, and most implementations of cable and fiber have no separation between "infrastructure provider" and "bandwidth provider".
> If you bring traffic directly to their local central office, they have to take it and deliver it up to the speed that they've sold to their customers.
What if their central office isn't connected to anyone else directly, because they get all their bandwidth and peering from one or more of the major backbones? Presumably they can say "go to those backbones", rather than having to start making peering arrangements with individual services.
> but decline and rely on peering (and install their CDN equipment across the street) if they try to charge monopoly prices
What if their central office doesn't have an "across the street" where you can put servers? Or what if they don't have a "central office", period?
I appreciate that you bring these questions up but your overall point seems to be "this is complicated and you can't provide logically / morally perfect arguments", which seems a little pointless.
All legal matters are complicated, but we're a modern society full of a lot of bright people and "free market lol ISPs get to be highway robbers now" doesn't seem like the best we can do.
I also disagree with the notion that inconveniencing some billionaires is less moral than hurting millions of consumers and setting a horrible precedent, so: regulate the shit out of them. If poor people can bootstrap huge medical bills then ISPs can handle some simple regulation.
e: you didn't make that point about prioritizing rich over poor etc, that was directed more broadly at net neutrality opponents
My argument isn't "this is complicated so let's not"; it's "this is complicated and needs to be handled with care, not with a trivial attempted fix like 'dumb pipes only' the way people often seem to think". Together with a healthy dose of "regulation has historically failed to adapt to future technologies". And we don't really want regulations that privilege the tech we currently know over the future tech we don't; that tends to lead in unpleasant directions, including regulatory capture and blocking of future competitors.
I think the fundamental problem here is that in most locations we don't have a market at all, free or otherwise. We're looking at all these restrictions in an effort to work around the fundamental problem of "people have little to no choice in ISPs". If you're going to have one ISP in an area and regulate its behavior heavily, you might as well just have the government run it. (Note: that's not a recommendation.)
I'd be entirely in favor of solving the problem by saying "last-mile infrastructure (from customers all the way to the nearest meet-me room) is a public utility, all ISPs have to remain separate from the last-mile infrastructure". If the average customer had a choice of half a dozen or more ISPs, none of these problems would arise, because any ISP stupid enough to make services deliberately worse would get dropped in favor of ISPs that don't. ISPs could then compete on everything from services provided, customer service, neutrality, geek-friendliness, IPv6 availability, CDNs/mirrors, or anything else they think people want.
Because many customers don't actually want a dumb pipe.
Dumb pipes don't come with email addresses.
Dumb pipes serve YouTube more slowly, and might not have the capacity for HD/4K video. (Look at what server the video comes from; if you're on a major ISP, the server name likely has your ISP's name in it.)
Dumb pipes serve Netflix more slowly.
Dumb pipes can't hold as many customers.
Dumb pipes might not get as many customers.
Dumb pipes cost more.
Personally, I'd like a dumb pipe only as far as the nearest major datacenter, and then quite a lot of smarts in that datacenter. I don't just want a raw connection to a major bandwidth provider, though.
See also the last paragraph of my previous reply: I'd be in favor of separating out the "dumb pipe" as a utility, and then having ISPs operate independently from that. I think that would solve the majority of the problems.
> Because many customers don't actually want a dumb pipe.
Yes, they do. That's exactly what they want. A dumb pipe that transports their bits and nothing more.
I'm omitting your list about dumb pipes, as nothing you write has anything to do with dumb pipes and is all completely wrong.
> Personally, I'd like a dumb pipe only as far as the nearest major datacenter, and then quite a lot of smarts in that datacenter. I don't just want a raw connection to a major bandwidth provider, though.
Hate to disappoint you, but there are no smarts in any ISP datacenter. It's just a big router and some interconnects. Raw bandwidth is all you get, if you are lucky.
> See also the last paragraph of my previous reply: I'd be in favor of separating out the "dumb pipe" as a utility, and then having ISPs operate independently from that. I think that would solve the majority of the problems.
It's not that simple, there are a lot of implementation details that matter. Do you want local loop unbundling, bitstream access or only resale?
I think we have a serious problem of definitions (which goes back to what I said about this being not trivial to define): we're both saying the phrase "dumb pipe" and apparently meaning quite different things by it. Defining terms differently does not invalidate arguments, it just means we're talking past each other.
> Do you want local loop unbundling, bitstream access or only resale?
In an ideal world? A form of local loop unbundling, with a not-for-profit entity running/maintaining the local fiber, and umpteen ISPs readily available to handle termination and routing at the other end.
> I think we have a serious problem of definitions (which goes back to what I said about this being not trivial to define): we're both saying the phrase "dumb pipe" and apparently meaning quite different things by it.
> In an ideal world? A form of local loop unbundling, with a not-for-profit entity running/maintaining the local fiber, and umpteen ISPs readily available to handle termination and routing at the other end.
That's nice. Now you've either locked out small/unsophisticated ISPs/network users who can't or can't afford to light the fiber or increased cost and/or complexity by requiring an additional service provider on top of the dark fiber provider to provide bitstream, retail and other transport services. Also excellent choice for finger pointing when these two actors donit get along.
Neither do rack servers. End customers don't buy the dumb pipe directly.
> Dumb pipes can't hold as many customers.
The same fiber can transfer the same amount of traffic.
> Dumb pipes cost more.
Cost more than paying monopoly prices for services?
> I'd be in favor of separating out the "dumb pipe" as a utility, and then having ISPs operate independently from that. I think that would solve the majority of the problems.
Sure it would. But it has to be one or the other. Would Comcast rather have network neutrality or divest its ISP/TV business?
> Neither do rack servers. End customers don't buy the dumb pipe directly.
Of course; they buy it from ISPs, and now you have the same problem again.
> The same fiber can transfer the same amount of traffic.
It doesn't matter how much your fiber can transfer if the arrangements at the meet-me room can't get you that much content for a price you want to pay.
> Cost more than paying monopoly prices for services?
For the same bandwidth? Yes. Raw bandwidth costs more than a plan from an ISP offering that much bandwidth.
> Would Comcast rather have network neutrality or divest its ISP/TV business?
I don't want a single choice of ISP, whether that ISP has agreed to network neutrality restrictions or to divest any particular services. I'd like a dozen ISPs to choose from, and then the rest stops causing a problem.
> Of course; they buy it from ISPs, and now you have the same problem again.
What problem?
> It doesn't matter how much your fiber can transfer if the arrangements at the meet-me room can't get you that much content for a price you want to pay.
So free isn't enough for you?
You can pretty much get as much content as you want, if you have the customer demand for it.
Join an IX and you'll get as much content as you want for free for the cost of the port. If you got serious content demand, do a private interconnect and get the content for the price of a crossconnect. There are facilities with no recurring costs for crossconnects. Need a shitload of content? The content providers will send you a CDN node to you, for free.
In summary, your your statement is provably false.
> Raw bandwidth costs more than a plan from an ISP offering that much bandwidth.
You obviously never bought wholesale bandwidth. Wholesale IP transit is 20 cents or less per Mbps. IX peering is cheaper than that. Private interconnects, again cheaper.
I don't see ISPs offering 100M connections for $20 or less.
>> Of course; they buy it from ISPs, and now you have the same problem again.
> What problem?
The message I originally replied to said "End customers don't buy the dumb pipe directly." And once an ISP starts trying to provide value above and beyond the pipe, all the same questions about neutrality come up again.
> If you got serious content demand, do a private interconnect and get the content for the price of a crossconnect.
> The content providers will send you a CDN node to you, for free.
What happened to "dumb pipe"? Both of these, especially the former, seem to contradict that.
> You obviously never bought wholesale bandwidth. Wholesale IP transit is 20 cents or less per Mbps. IX peering is cheaper than that. Private interconnects, again cheaper.
I have, in fact, looked into both wholesale bandwidth and IX peering (though not in the last year or two). Both of which, at least in the areas and exchanges I've looked in, were several times higher than the cost you're quoting. It's entirely possible that you have access to better information/markets, or that ISPs can get much lower pricing; I based the comment I made on the information I had.
All the numbers I've seen suggested that raw bandwidth costs significantly more than ISPs charge for it, because consumer ISPs rarely expect 24/7 utilization of the full bandwidth, and likely also because they can provide some of the larger sources of content for free out of their own datacenter.
> The message I originally replied to said "End customers don't buy the dumb pipe directly."
But they do, and that's what they want to keep on doing. An Internet connection is by definition a dumb pipe. It's only when you start messing with it and prioritizing traffic that it becomes something else.
> And once an ISP starts trying to provide value above and beyond the pipe, all the same questions about neutrality come up again.
No, this is just a silly argument. Just because an ISP provider other services in addition to or on top of the Internet connection does not make nor any net neutrality issues. It is only when they start prioritizing stuff that there is an issue.
> What happened to "dumb pipe"? Both of these, especially the former, seem to contradict that.
Nothing. Seriously, this is how the Internet works and is supposed to work. You are supposed to interconnect networks, that's why it's called the Internet. Doh!
Just because somebody else has a shorter path to you than somebody else doesn't make it wrong or unfair.
> All the numbers I've seen suggested that raw bandwidth costs significantly more than ISPs charge for it, because consumer ISPs rarely expect 24/7 utilization of the full bandwidth, and likely also because they can provide some of the larger sources of content for free out of their own datacenter.
Well, it doesn't and it's not for the reasons you state. Wholesale bandwidth prices decrease by 30% on average year on year. Consumer prices are stagnant, or rising.
While it's true that consumers don't use their connection 24/7 (although some make a pretty good go at it), the ISPs could without much problem support that use case. Furthermore ISPs provide close to zero of the content consumers consume. All the content is coming from the content providers, either directly, via an IX or a local cache. The ISPs have nothing to do with it, they are just dumb pipes.
It means you control a physical data wire that goes into more than a thousand residential buildings.
> DSL is no longer competitive, and most implementations of cable and fiber have no separation between "infrastructure provider" and "bandwidth provider".
You're the last mile ISP if you're the infrastructure provider regardless of whether you're also the bandwidth provider.
> What if their central office isn't connected to anyone else directly, because they get all their bandwidth and peering from one or more of the major backbones? Presumably they can say "go to those backbones", rather than having to start making peering arrangements with individual services.
Outside of megaplayers like Netflix and Google who are effectively Tier 1 networks unto themselves, the backbones and CDNs are the people who use the free peering. Some website like Yelp or Expedia doesn't want to peer with ISPs or run a CDN, they want to pay Akamai et al to take care of that.
> What if their central office doesn't have an "across the street" where you can put servers?
You can put them wherever you want to. Nobody stops you from putting all your servers for the whole world in one city, or having regional hubs, or doing whatever you want. That's your network, not the last mile ISP's. Level 3 can run fiber from an office 200 miles away if they think that's worth it over buying closer office space.
> Or what if they don't have a "central office", period?
Then where do they normally peer with other networks or operate CDN hardware?
> If you phrase it as just "don't prioritize traffic", you've just broken most CDNs, as well as QoS that optimizes for latency/bandwidth to improve VoIP and similar.
This is simply incorrect. CDNs are not prioritized on any ISP network. CDNs work because they have a shorter route to the end user, not because they have a higher priority.
ISPs should not and do not apply QoS in their core for VoIP or anything else when it's part of generic Internet traffic. It's the end users responsibility to QoS police VoIP or other traffic on their WAN interface. The ISP just shifts the bits.
Not privatizing traffic does not break CDNs nor customer QoS. Neither of these things have anything to do with net neutrality.
> But even then, how do you draw a bright line between "deliberately degrading", "not giving an advantage to", and "not doing anything to address bandwidth problems when they arise"?
It's pretty obvious how you draw that bright line. Don't prioritize and don't congest interfaces.
> How would you suggest writing a regulation that doesn't break CDNs and similar, and that can tell the difference between "deliberately hurting" and "not helping"?
Easy. As I already wrote, forbid prioritizing and saturation of network interconnects.
> How will that regulation adapt when people come up with new ideas to make the web faster?
It does not need to when done right, as explained above.
> One angle I could imagine working: limiting the ability of one company or family of companies to run both an ISP and network services via that ISP. Still hard to write something robust, though.)
Nothing hard about it. It's called functional separation and it's a widely applied remedy to the anticompetitive nature of telecom markets in other countries.
> CDNs are not prioritized on any ISP network. CDNs work because they have a shorter route to the end user, not because they have a higher priority.
So, anyone who can successfully define what they're doing as "not prioritizing" can entirely sidestep any regulation you have in mind.
Putting a server in the ISP's datacenter is a common approach. Traffic to those servers doesn't have to go over any external peering links.
> ISPs should not and do not apply QoS in their core
Many do. As one simple example, look at the ISPs that accelerate the first N bytes of connections, to prioritize interactive requests over bulk transfers/downloads. Other ISPs apply more sophisticated approaches.
> It's pretty obvious how you draw that bright line. Don't prioritize and don't congest interfaces.
> forbid prioritizing and saturation of network interconnects
So, receive and deliver all traffic, regardless of available capacity or financial constraints? Provision every part of the network so it can handle all the traffic passing over any other part, because they're not allowed to limit bandwidth in any way at any point?
A saturated interconnect might mean "we're trying to screw Netflix", or it might mean "we can't afford it right now" (either because they're a small ISP and don't have the money, or because they're spending it on something their customers want more).
> So, anyone who can successfully define what they're doing as "not prioritizing" can entirely sidestep any regulation you have in mind.
This argument makes no sense. Colocating a CDN node has nothing to do with prioritization. Nobody is fiddling with any QoS settings.
> Putting a server in the ISP's datacenter is a common approach. Traffic to those servers doesn't have to go over any external peering links.
What is your point?
> As one simple example, look at the ISPs that accelerate the first N bytes of connections, to prioritize interactive requests over bulk transfers/downloads.
This is an incorrect characterization of PowerBoost. PowerBoost uncaps your DOCSIS connection temporarily to take advantage of any momentarily free spare capacity. This is not prioritization, it's getting a free tier upgrade temporarily.
> Other ISPs apply more sophisticated approaches.
Such as?
> So, receive and deliver all traffic, regardless of available capacity or financial constraints?
Don't be silly. As long as the ISP provides whatever speed tier they sold to the consumer all is good and well. If they cannot do that then they are shortchanging the consumer and should be penalized for that.
> Provision every part of the network so it can handle all the traffic passing over any other part, because they're not allowed to limit bandwidth in any way at any point?
Yes, to the extent required to provide the service they sold.
> A saturated interconnect might mean "we're trying to screw Netflix", or it might mean "we can't afford it right now" (either because they're a small ISP and don't have the money, or because they're spending it on something their customers want more).
If you saturate your interconnect you fail at ISPing. There is no excuse. Interconnect capacity is cheap and plentiful.
I think the CDN fix could be built right into net-neutrality laws: If you're going to support that functionality then also do it equally to any CDN that is also willing to foot the bill and/or meets certain criteria. For example, don't only let Netflix's smaller competitors let that box sit on your network.
(No, this regulation wouldn't be burdensome either.)
QoS is something I'm unfamiliar with when it comes to ISPs. I personally wouldn't be offended if some small % of bandwidth was dedicated to services consumers and businesses use that require priority, such as VoIP. As long as that applied to all VoIP services and apps equally, though I'm not sure how you would be able to tell.
> As long as that applied to all VoIP services and apps equally, though I'm not sure how you would be able to tell.
The only fair way to do this is to use port numbers and protocols and endpoint IP addresses only for the purpose of telling which packets belong to which traffic flow or customer. For deciding which flows get priority, you should look only at the actual pattern of traffic: sparse low-rate flows should be assumed to be interactive, in the same manner that your operating system's CPU scheduler assumes that processes with intermittent load are more latency-sensitive than long-running sustained loads. To see how this kind of packet scheduling works in the real world, look at the fq_codel and cake queue disciplines in Linux. The former has become the default for many Linux distributions and is used by some of the more modern commercial wireless routers.
This will have the side effect that my ssh session will get the same preferential treatment as your VoIP, unless I'm using it to tunnel rsync.
There's absolutely nothing wrong with, for instance, a content provider handing an ISP a box that sits on their network, mirrors/caches content, and serves it up to that ISP's customers. That helps the content provider, the ISP, and the ISP's customers. And I don't think the majority of "net neutrality" advocates want to break that.
When people talk about "net neutrality", the problems they have in mind tend to look like either an ISP extorting content providers to extract extra revenue, or an ISP giving their own first-party services an advantage by deliberately degrading third-party services. For instance, see Comcast and Netflix.
But even then, how do you draw a bright line between "deliberately degrading", "not giving an advantage to", and "not doing anything to address bandwidth problems when they arise"? (Comcast could claim the latter regarding Netflix, even though it seems obvious what it means when they don't arrange additional bandwidth on those saturated links.) What if the ISP provides the CDN infrastructure, rather than the content provider? What if the ISP contracts with a third-party CDN?
How would you suggest writing a regulation that doesn't break CDNs and similar, and that can tell the difference between "deliberately hurting" and "not helping"? How will that regulation adapt when people come up with new ideas to make the web faster?
(One angle I could imagine working: limiting the ability of one company or family of companies to run both an ISP and network services via that ISP. Still hard to write something robust, though.)