Hacker News Reader

Comment by wtallis | Hacker News Reader

> How can an ISP get Netflix to pay it, or lean on Netflix to install new servers in their network, if it is illegal to use their own bargaining position?

ISPs don't have to lean on Netflix to install CDN nodes within the ISP's network. If Netflix finds that it's cheaper for them to do so rather than buy transit to the last-mile ISPs from Netflix's existing ISPs, then Netflix will come to those ISP's offering money and begging for the rack space. If, on the other hand, it turns out that Netflix doesn't need the low latency that CDN nodes afford and that they can more cheaply obtain the necessary bandwidth indirectly through Level3, etc., then the last-mile ISPs should not be empowered to distort that market outcome.

View on HN

So lets say the ISP does not have a business relationship with either Netflix (because its illegal under net-neutrality) or with Netflix's existing ISPs because there are others in between. But the path used to route movie packets is expensive. How does the ISP lean on Netflix to pay for its part of their operating costs?

> How does the ISP lean on Netflix to pay for its part of their operating costs?

They don't. It's purely your accounting fiction that is attributing those costs to Netflix instead of to the ISP's customers who are using Netflix.

If streaming video increases the ISP's operating costs, they can simply increase prices, preferably through a fair metered usage model that can address all of the video streaming usage costs together, instead of just the Netflix costs.

So if data from company A costs the ISP more that the same quantity of data from company B to get a hold of, how ultimately should this cost be passed on?

Separate metering for end users? Targeted agreements with large hosts? Same metering for all, screw those who are frugal with packets.

> So if data from company A costs the ISP more that the same quantity of data from company B to get a hold of,

Expand that hypothetical a bit more, so that we can determine if the situation you're asking about is one that ever happens in the real world. Are you asking about traffic imbalances at peering points? If so, the agreements in those cases are between ISPs and do not include end users like Netflix as a party to the negotiations or payments. The primary purpose of an ISP is to ensure that their subscribers don't have to worry about whose lines their packets might travel on to get to their destination.

Not traffic imbalances at peering points. Rather a situation where data has to travel further, through more nodes from source to destination. Shouldn't a logical system allllow that data to cost more? Or for example through the same number of nodes but more expensive source nodes.

Beyond that, there is some "discrimination" I would be just fine with. For example why should the end consumer pay the cost of YouTube ad data?

The fear is of course discrimination on political or ideological grounds, not just economic. but it would be fairly obvious and nearly impossible technically for an isp to pull that off. And a net neutrality regulation or even ownership at the state or local level would be perfect. You only need net neutrality at the end points where there is no competition, not in the network at large

> Rather a situation where data has to travel further, through more nodes from source to destination. Shouldn't a logical system allllow that data to cost more?

That's not what the cost structure looks like for ISPs, so you would need a pretty good justification for making the price structure look like that. It doesn't necessarily cost the ISP more to have packets travel across more hops within their network; it's more likely to reduce the ISP's expenses by reducing the amount of traffic they have to buy transit for. Where a packet is within the network really only matters and becomes worth tracking when a link gets congested.

" Same metering for all, screw those who are frugal with packets."

Metering would be by bandwidth used, so those who are frugal with packets would inherently pay less.

Your grasp of the concepts seems confused.

"lets say the ISP does not have a business relationship with either Netflix (because its illegal under net-neutrality)"

That is not illegal under net neutrality, and it is basically irrelevant. Netflix, like any Internet user (yes, that is what Netflix is) must contract with some ISP (probably many ISPs given their scale). Net neutrality does not prevent ISPs from selling rack space in their data centers and allowing companies to colocate their servers with the ISP's routers.

Net neutrality is really about the logic ISPs use for routing, and the basic requirement is that routing decisions should not be based on the sender address or the application. There is plenty of room for argument about how strictly that should be enforced e.g. some experts support different priority classes for different application types while others believe in a stricter form of net neutrality. The general idea is that the performance of Internet applications should depend solely on technical details and that ISPs should not be allowed to impose artificial or arbitrary restrictions. There are plenty of legitimate questions over what is truly a technical reality and what is artificially imposed by an ISP (e.g. if an ISP uses the same physical infrastructure for Internet and non-Internet services, is it artificially constraining the Internet users by denying them capacity that is not being used for the non-Internet services?) but for colocation there is no real debate and nobody is claiming that is a NN violation.

> some experts support different priority classes for different application types

That's a thoroughly outdated QoS paradigm. It's almost impossible to make rule-based traffic classification neutral and fair, because it always privileges existing applications and protocols over upstarts that aren't correctly identified by the ruleset. It's also far too easy for ISPs to "forget" to properly classify their competitor's traffic when the ISP is also a content provider, and poorly-designed rulesets can be gamed by using protocols/applications on non-standard port numbers.

Fortunately, this QoS paradigm is no longer necessary and the state of the art for QoS has moved on to techniques that only need to look at the quantity and size of packets in order to correctly infer the correct latency vs throughput tradeoffs for each traffic flow. Packet scheduling is now much more like CPU scheduling, in that it works well enough out of the box without requiring manual tuning or prioritization.

Reply on news.ycombinator.com