That's a mischaracterization of the situation.

ISPs complained about the amount of traffic originating from Netflix.

This traffic was requested by the ISP's customers, and the bandwidth to carry it was promised by the ISP.

But, still, the ISPs were unhappy. So Netflix approached them and said they are willing to offer local caches of Netflix content, for free, to any ISP that requested it to reduce network congestion.

This was a gesture of good will by Netflix, not an attempt to outsource their hosting costs. It is, fiscally, a loss to Netflix because they need to purchase and maintain the storage appliances for open connect partners.

Actually I don't think the fiscal loss is so clear-cut. Each deployment of cache nodes removes load/bandwidth from the rest of the CDN, and also reduces support time spent with customers.

In fact, 100% of Netflix traffic outside the US and 90% inside the US is served from Open Connect[0]. It is actually a shift of the location of the serving hardware that allows both parties (the ISP and Netflix) to save on bandwidth. This is a net win for everyone.

[0] http://arstechnica.com/information-technology/2014/05/netfli...

Because your homepage doesn't attract many TB of traffic?

So, bandwidth. What is the threshold with which ISPs should start giving out free hosting?

It's not free hosting. It's not even hosting a webpage because it's not internet facing at all. Comcast customers can't visit the "free webpage" hosted on Verizon because it's not the internet we're discussing here. It's delivering a service to the local ISP network subscribers.

The ISP is allowing Netflix to bring caches and servers to their interconnect so that the ISP can meet the demands of its users who are requesting that data.

When should an ISP give out "free" things? When the quality of service for their customers requests depends on it.

It's that simple: Customers request data, and ISPs have an obligation to deliver that data once it arrives at their local network.

What you call "Free hosting" and "free upgrades", I call "upgrades for requested service paid for by my monthly bill".

Because when I request data, and that data requires infrastructure to reach me adequately, then I am paying for that infrastructure by virtue of my monthly bill, and my decision to use part of my capped bandwidth on that service.

Remember: residential internet is capped because "users must pay for their use, and infrastructure can't handle too much demand so we need to charge users to upgrade according to their usage pattern". So therefore, I'm paying for that Netflix interconnect, it's not free, I pay for it, because I pay for 300GB of data a month and all of the necessary infrastructure to deliver it!

Econ 101: all things being equal cheaper is better than more expensive.

If I have a network and my customers use Netflix and aren't going to not use Netflix because I don't want them to, then part of my job is to ensure that my customers get the bandwidth I've promised them. (This isn't how it ACTUALLY works but how it's SUPPOSED to work)

As an ISP I have really rather large amounts of bandwidth in the last-mile at least in aggregate. Let's say that I can reasonably offer 20Mbps to each of my 1mm customers from my POPs to their houses. That's 20Tbps in aggregate. I probably don't have 20Tbps worth of back-haul from all my POPs to all the peering stations where I actually get the customers connected to the internet at large.

If network traffic is all long-tailed and the biggest use of bandwidth is 1% of capacity and it goes down from there "free hosting" doesn't make sense. But what if traffic to one company makes up 30% (or 80%) of total back-haul utilization at peak hours? I'm spending a lot of capacity for a single destination.

Now what if that place offered to create a magical wormhole from their servers to my customers at my POPs such that a large fraction -- say 80% -- of my customer's traffic from/to them never hits my back-haul it just appears out of thin air at the POP. Would I consider this a good deal? Depends on how much I pay for the back-haul versus how much electricity they're going to use at the POP. All-in I would suspect that it is a good deal thinking in these terms.

That's precisely what the Netflix appliance is. It's a way to give the customers Netflix without costing any bandwidth on the back-haul network that ISPs operate.

The reason that ISPs aren't all jumping right on this (despite the likely cost-savings) is that they view Netflix as the competition and they're prefer an adversarial relationship that hopefully puts Netflix out of business rather than cooperating and in their minds speeding their own demise.

So the rule of thumb you're looking for is that ISPs should start giving out free hosting when it's cheaper to give free hosting than to pay for the back-haul.

> rather than cooperating and in their minds speeding their own demise

It makes sense, then, that one of the few large cable companies that uses Open Connect is Cablevision, who's CEO is on the record saying "Ultimately over the long term I think that the whole video product is eventually going to go to the Internet."[1]

He's one of a few that has accepted the eventual fate of cable TV, and so his business decisions aren't biased by a need to delay the inevitable.

[1]: http://online.wsj.com/news/articles/SB1000142412788732342060...

That's a nice theory, but the impression I get from the (extra biased against last mile ISPs) DSLReports is that Comcast is continually spreading usage caps and plans to make them nationwide in 5 years, e.g. http://www.dslreports.com/shownews/128987

Lots more with this search: http://www.dslreports.com/nsearch?cat=news&q=comcast%20caps

There is no inherent problem with usage capped plans, as long as the price is fair. If you'd like an uncapped plan, pay more.

The problem is that the last-mile providers try to extract money from netflix for a service that the consumer already paid for (deliver those video bytes)

You don't see a problem with e.g. Comcast implicitly charging the consumer extra for using Netflix (by exceeding the cap and paying by the drink afterwords) and not charging anything "extra" for using Comcast's video services?

There is an argument there, in that it costs more real money to deliver video bits the a la carte Netflix way than the cableco broadcast way, but I believe the conflict of interest remains an issue, and the prices I see for exceeding these rather small caps don't strike me as fair.

(Albeit AT&T's, the only choice I have aside from a not so reliable WISP, are particularly ridiculous: 150 GiB/month including I'm not sure what overhead for a continually rising price 2nd from the bottom "up to" 1.5 Mbs down/300+ Kbs up line that currently costs $36/month, each additional 50 GiB costs $10. We'd get a faster line so my father could watch video, at $5/month extra each increment, if the cap wasn't so low and the overages so high.).

That's not a capped plan. A capped plan charges you more if you exceed a specific bandwidth threshold - no matter what you spend the bits on. As long as Netflix, Comcast PPV, Youtube and my private cat videos get charged the same price - I'm fine with that.

The issue you're pointing out is the lack of net neutrality - but that is orthogonal to the pricing.

Comcast, yes. Cablevision, no. I was speaking of the latter.

Oops!

So you were, sorry about that.

When it's cheaper for them to host the caches a service offers than increasing their peering bandwidth. That's a fairly simple exercise for controlling.

When the space and electricity for the hosting is cheaper than the transit costs they would otherwise incur.

It's not that they're providing free hosting. Netflix is providing hardware that ISPs can install in their own network locations (which they typically have plenty of). In exchange, their customers get better quality and ISPs don't spend their uplinks on Netflix content. It's pretty win-win for everyone.

Your question is backwards. ISPs can choose this option to save themselves money.

I suspect that the real reason that ISPs like Verizon are resisting, is that with open connect in place there would be nowhere left to hide and blame the internal ISP bandwidth problems.

It's not hosting. These devices are not available to the open internet, only customers of the ISP who's network they are on.

And I'd say the threshold is the point at which you are no longer to deliver the content by other means.

First, you'd have to buy and build a server for each ISP containing that content and support it when things go wrong. Next, it would have to be cheaper for the ISP to host your server on their network than for them to upgrade their connections to the backbone to handle traffic to your site.

If those two conditions are true, I am sure ISPs will accept your offer.

Possibly because your homepage does not drive a third of Internet traffic, both from within the ISPs network and from without.

The Internet is a collection of public and private networks. When it costs more to upgrade the linkages between your network and everyone else's than to host a server inside it, that is when the ISP should pay to support a local resource.

You're comparing apples and oranges, really. If your website ever reaches the point where traffic starts to saturate an ISP's connections, I'm sure they would be willing to discuss a local CDN resource with you, too. Whether they offer to pay for that or charge you for it depends largely upon how stupid and/or greedy their management is.