Comment by danudey | Hacker News Reader

Comment by danudey | Hacker News Reader

Parent Full thread

The problem, as I see it, is that Verizon/Comcast (and other ISPs) are overselling, and the amount of traffic that Netflix is generating is showing that. Verizon sells X customers Y-bandwidth connections, but they don't have X*Y worth of uplinks, so they can't supply data to the people who want it.

Until now, this hasn't been a problem, because most people didn't max out their connections, and if they did they didn't do it for very long. The very nature of Netflix's business, though, is to serve the best traffic that the customer's connection can handle, and that starts to add up. Unfortunately, it adds up so much that in some cases (e.g. Comcast) you can't even serve VHS-quality content to all the customers who want it.

Because Verizon is overselling their bandwidth, they're now having issues where they can't service the demand that they sold their customers on. Instead of upgrading their upstream connections (which would cost them money, which they have a lot of), they want Netflix to pay for it. The logic is that Verizon wouldn't have these problems if Netflix weren't sending so much content; that argument breaks down when you consider that Verizon sold their services based on available bandwidth and now can't provide it; Netflix is just the service that exposed the problem.

Netflix has other solutions, though; they can provide servers that get hosted in the ISP's network to provide content to users without traffic coming from outside the ISPs' networks (thus freeing up their normal uplinks for regular traffic), or by peering with Netflix at various points (which accomplishes the same thing).

Unfortunately, neither of these solutions involves Netflix giving money to Verizon, which is why Verizon doesn't want to bother with going to the time and trouble when they can force Netflix to pay for their infrastructure upgrades instead.

All ISPs oversell; the two alternatives are (a) really low bandwidth for everyone, or (b) really expensive bandwidth with lots of wasted capacity that people seldom use. If you're arguing against contention, you're arguing for a massively inefficient allocation of money. This is basically a thin provisioning vs thick provisioning argument. Thin provisioning makes lots of sense when not everyone is using the maximum all the time, and especially when thick provisioning is expensive. Uncontended bandwidth is expensive.

If customers who watch Netflix are using more bandwidth on average than customers who don't watch Netflix, who should pay more?

If you think Netflix shouldn't pay more for their bandwidth (and then take it out of their customer), you're arguing that Verizon should pay for the extra bandwidth (and thus take it out of their customer).

The logical conclusion is that Netflix-watching users would be subsidized by non-Netflix-watching users.

Alternatives might be a special "Netflix" package sold by Verizon that has better quality interconnects to Netflix. Would you agree with that? Not very net-neutral though.

The statement that all ISP's over-sell is oversimplified.

All ISPs statistically multiplex the bandwidth requirements for off-net connectivity. Over-selling a peering relationship is impossible, as the ISP is selling only a speed-rated connection to it's own network. The assumption (or contractual service level) is generally that the ISP will augment either transit or peering as the resource becomes saturated.

The issue on hand is not the local or last mile bandwidth between the (retail) ISP and the customer, but the interconnection between the retail and the wholesale bandwidth providers. Two of the wholesale providers in question have publicly stated that their interconnections with specific retail providers are congested, and are not being augmented per their private contracts with the retail networks.

The above explanation is also a simplification, because for example Verizon is also a premium wholesale network in North America and several MSOs have also been building national transport networks.

The logical conclusion is that Netflix-watching users would be subsidized by non-Netflix-watching users.

Verizon chose that business model; we're just asking them to stick to it. Or they could meter. But charging both sides is bad for innovation.

If we're going with metering, I hope cable television watchers can join in the fun. If I subscribe for cable television and I only watch two hours a month, why should I subsidize someone who uses it 12 hours a day?

The argument based on costs is that it costs a cableco, or e.g. satellite company, no additional incremental cost to serve you and the 12 hours a day customer, all bits for all channels are broadcast to all customers.

Whereas each additional outside their net a la carte video stream does cost, if it's at a period of peak demand. The build out for peak demand is what costs, well, everyone, including the last mile ISPs, and video does tend to coincide with or now probably creates peak demand in evenings.

Some technical solutions like pulling during off peak and storing it for later viewing are unacceptable to content providers :-(, then again, none of this matters in a good way to a last mile company who very much wants you paying them for video rather than Netflix et. al., which is all of the big US ones except Century (formerly US Worst).

Interestingly, the last time I checked, a few years ago, one of CableOne's standard capping mechanisms was only for usage at peak periods....

> The argument based on costs is that it costs a cableco, or e.g. satellite company, no additional incremental cost to serve you and the 12 hours a day customer, all bits for all channels are broadcast to all customers.

Doesn't matter. The person watching 12 hours a day is getting a far greater value from their subscription than I am, which is the same reason the (generally all-encompassing, as opposed to peak metering) caps are being argued as fair, because of the value received.

Now of course all the ISPs policies make sense when you assume their motives are to kneecap a competitor and not to fairly price their service.

Or maybe the opposite ?

If you charged Netflix, they would consider investing more money on trying to reduce the bandwidth they use: - innovative video codec - better encoders - new streaming technology

(Ignoring sunk costs:)

The ongoing costs to an ISP are not the costs of providing bits. Let me hand-wave a bit and say the 99th percentile of families is using 4 hours of Netflix per day, and that's about 1 GB per hour, so 120GB per month, which is maybe a few bucks of costs to the ISP.

The ongoing costs to an ISP are constantly building out and upgrading the network so that when we want to watch Netflix on Tuesday of next year at the peak hour we can. If there is metering or preference used, it should only apply at the peak times.

The problem isn't that they're overselling per se, the problem is the deception involved. They sill this "up to x megs" service, with the knowledge of that overselling in mind, they can't ever provide that speed. In effect, selling you "up to" some amount of bandwidth when it is mathematically impossible for you to get that amount, ever.

False advertising.

In a perfect world: I'd like to see a "broadband facts" that ISPs are required to provide, including average single-subscriber connection throughput to the their subscriber's top 100 sites, equivalent packet loss and jitter, amount of oversell (combined max subscriber rates vs uplink/downlink capacity). This data must be posted, prominently, on all sales literature and sites, and communicated verbally over phone. It must be updated no less than once per month.

Reply on news.ycombinator.com