An end-user ISP is going to have a lot of different pipes for a lot of different transit providers, which are going to have unequal utilization and different data. Internet bandwidth is not fungible in the way that, say, electricity is.
Each network provider that the packet is passed thru inspects the destination address in order to determine which pipe to send it on to get it closer to its destination. If that pipe is congested, the packet can't turn around and find another route. It has to be dropped and retransmitted. This is true for streaming video packets as well as others like email, video game, etc (although most video games use UDP which cannot retransmit, if it's dropped, the game jitters)
So if I'm an ISP downstream from a congested pipe that happens to be transiting a lot of streaming video, I know that all of the users of that pipe who are not streaming video are having a bad experience - slow text web page loads, jittery video games, garbled voip calls, etc. So what I can do is selectively drop some of packets from the video streams (I can tell which ones the heavy users are if I record the source IP and count the # of packets per stream), and those streams will generally gracefully downgrade to the next lower streaming quality. This relieves the inbound congestion because the sender stops sending as much data to users downstream of me, and allows my other customers to play video games without lag, make voip calls, load web sites without long delays, etc.
So clearly the pipe needs to be upgraded, the question is, who should pay to have that pipe upgraded? That's between the ISP and the transit partner, they have peering agreements to determine how that cost is shared. I think it's important to look into who failed to live up to their contractual obligation to shoulder the shared costs.
BTW - going thru a VPN on Verizon just means that the traffic is getting to the customer through a different upstream provider than the one that BGP specifies because it originates from a different network, one that isn't as congested as the BGP route from Netflix. You would need to look at a traceroute to figure it out.
In Comcast’s case, the congested pipe was just the connection between Level3’s network and Comcast's. Level3 and Comcast had an existing settlement-free peering agreement (no money paid in either direction), and Level3 wanted to keep it that way, but Comcast wanted to start charging Level3, arguing that it was sending it far more traffic than it was receiving. Level3 argued that it shouldn’t have to pay because (most of) that traffic had been requested by Comcast customers in the first place.
Also:
> [Level3] tells me that it actually offered to give the necessary hardware to Comcast (at around $50,000 per port) and that it offered to do "cold-potato" routing deep into Comcast's network, dropping off streaming traffic near the customers who requested it, for instance.
Thanks, that article completely illustrates what I'm talking about.
> Who paid for the delivery of all this on-net traffic, then? The customers. In Level 3's case, this means that CDN customers like Netflix would pay Level 3, while Comcast's cable modem subscribers would pay Comcast. Very simple, very clean, and according to Level 3 now, this is the way the Internet should be connected.
> But after winning the Netflix deal this autumn, Level 3 suddenly wanted to pass far more traffic over its links with Comcast. Comcast balked; Level 3 suddenly looked less like a transit vendor and more like a CDN.
You are correct that they had an existing settlement-free peering agreement. But that agreement had traffic ratio requirements which Level3 broke because of its Netflix contract. So they can offer whatever else in the world that they want to seem like "nice guys" - free routers or cold potato routing or whatever. But Comcast is not obligated to allow Level3 to break the terms of their agreement so Level3 can make tons of money off Netflix at Comcast's expense.
What you say is true. But the counterarguments are:
- At least according to Level3, it was not standard practice in the industry to charge for peering (not transit - the final destination for the packets was in Comcast's network), regardless of traffic ratios.
- You could say that it should be standard practice, but that would be unfortunate because Comcast effectively has a monopoly on fast connections to its own customers. Since it's much easier for consumers to switch video services than ISPs, failing to reach a deal would be much worse for Netflix/Level3 than for Comcast - so there would be little preventing Comcast from charging as much as it wanted.
This was more true in 2010 when Netflix hadn't yet started creating original series. These days it arguably has more leverage due to being the exclusive provider of those series... but it would still be a huge risk, a huge money-loser for them to walk away from Comcast, or any other ISP, and get into a cable-blackout-like situation. Anyway, cable blackouts feel distinctly unlike how the 'open' Internet is supposed to work. Netflix is one thing, but what about, say, would-be competitors to Netflix that currently have few customers and almost no leverage?
- It would also arguably be double charging - charging consumers and backbone providers for the same packets. If you buy "Internet access" with 100Mbps download speed, but that actually means "100Mbps down from companies that pay the toll", are you really getting what you paid for? Admittedly, transit providers have similar practices, but the level of competition for transit makes that less of an issue.
- Also, most Comcast end-user connections have much higher download speed than upload speed, so even if users were saturating their (paid-for) connections in both directions talking to Level3, Level3 would still be sending more data than it received. In other words, rules about sending and receiving similar amounts of data never really made sense for peering between a consumer ISP and a backbone provider, even if this was less visible in the past.
Yes, we could argue back and forth about what is fair and what is standard. But what did the contract, agreed to by both parties, say? Was Level3 in the wrong to sell Netflix a product that they could not deliver without violating their prior contract with Comcast?
I agree it sucks to be a consumer and get shafted by these shitty deals, but let's at least be honest about how we got here and what the powers that be are actually gunning for.
I've said in other places in the thread that I believe that the big content companies (the ones who don't also own telcos) are trying to gain negotiating power over telcos/end-user ISPs in order to pay less for high-bandwidth use cases like video streaming. I don't believe for a nanosecond that this has anything to do with free speech, given the censorship track record from some of the biggest net neutrality supporters. I also don't buy the argument that end user ISPs will be able to charge extra for access to different content providers (although I think consumers will continue to gladly accept cheaper/zero rated content from some providers, in violation of net neutrality) - simply because while last mile competition isn't great, it does exist (especially in wireless), and there are enough regional ISPs to really put a dent in the big guys should they engage in a business practice like that.
Each network provider that the packet is passed thru inspects the destination address in order to determine which pipe to send it on to get it closer to its destination. If that pipe is congested, the packet can't turn around and find another route. It has to be dropped and retransmitted. This is true for streaming video packets as well as others like email, video game, etc (although most video games use UDP which cannot retransmit, if it's dropped, the game jitters)
So if I'm an ISP downstream from a congested pipe that happens to be transiting a lot of streaming video, I know that all of the users of that pipe who are not streaming video are having a bad experience - slow text web page loads, jittery video games, garbled voip calls, etc. So what I can do is selectively drop some of packets from the video streams (I can tell which ones the heavy users are if I record the source IP and count the # of packets per stream), and those streams will generally gracefully downgrade to the next lower streaming quality. This relieves the inbound congestion because the sender stops sending as much data to users downstream of me, and allows my other customers to play video games without lag, make voip calls, load web sites without long delays, etc.
So clearly the pipe needs to be upgraded, the question is, who should pay to have that pipe upgraded? That's between the ISP and the transit partner, they have peering agreements to determine how that cost is shared. I think it's important to look into who failed to live up to their contractual obligation to shoulder the shared costs.
BTW - going thru a VPN on Verizon just means that the traffic is getting to the customer through a different upstream provider than the one that BGP specifies because it originates from a different network, one that isn't as congested as the BGP route from Netflix. You would need to look at a traceroute to figure it out.