Neutrality of the Net

Certainly you know better than this. HTTP 1.0 had a severe performance problem because it opened a new TCP connection for every object. TCP applies a "slow start" algorithm to each new connection, which is a traffic control kludge intended to keep interior links from being overwhelmed. It was necessary to alter the design of HTTP to re-use TCP connections in order to boost its performance.

You're looking at things from the wrong end...the claim was that the network (devices between the two end points) didn't "look inside" IP packets. HTTP is an application level protocol, and there is no claim made by the author that either of the two end points is expected to remain transport-layer agnostic (which most network architects expect out of application level protocols though).

Your second comment addressed it better. Routers do look at the headers of IP packets, to determine their destinations, to be able to route them. So the original post was indeed inaccurate. It was unfortunate that you used an irrelevant argument in your first comment on it, it weakened your (valid) argument that the network does look inside packets.

"It's no secret that all the major innovations in communication and networking for the last ten years have taken place off the Internet: wireless phones, wireless LANs, location-based services, home networks, broadband over powerline, and location-based services. Opponents of "net neutrality" simply want to update the Internet's service space so that it can act as a faithful intermediary for these advanced networks and their QoS requirements. Do we need government's permission to do that, or can we and our investors conduct this experiment on our own?"

Lets use an analogy to look at what you're saying. The connectivity ISPs provide is like a toll road. You pay to use the road, either on a per-trip basis (pay per use), or you get a season pass (flat rate fee). The bandwidth an individual user uses can be compared to the kinds of / numbers of her/his vehicles allowed on the road for the price paid.

So you want to be able to charge
[1] Users who use that road and *need* preferential access - their vehicles would be useless to them otherwise.
[2] Destinations on that road which have users who *need* preferential access - charging them would be allowing them to subsidize their users, or pay bulk rates for their users and charge their users lower rates per-user, or give them special offers etc.

Obviously, the preferential users will end up being the ISP's most profitable customers since your costs are one-time investments.

So the questions that come up are
[1] What is the incentive to invest in maintaining the "normal" road, the road on which normal users travel, the ones who pay the existing fees? What is the incentive to improve the quality of service for this class of users as the technology becomes available to do that ? (e.g. IPv6 migration). The less maintenance that is done on these roads, the more pressure on destinations by these users to pay for preferential access. And users will end up paying more ultimately for those destinations' services, directly or indirectly.

[2] Do you not see a strong incentive for an ISP to degrade the quality of service to individual destinations, to incentivise them to move to a preferential access structure?

What the opponents of net neutrality want to do would work well in an ideal world with absolutely free markets and enough competition for every kind of big ISP to exist. That isn't the case, how many *huge* ISPs do we have? All that it takes is their forming an informal, unspoken cartel, because they see that its in their best interests to move to a preferential traffic structure. In that case, who protects the interests of the consumer?

Nobody. Thats where the government needs to step in. And we know that regulation when done needs to be done in the simplest fashion possible - the more complex the regulatory structure, the tougher it is to do business.

So the government needs to "define" its protection of consumers in the simplest fashion possible...to offer businesses including ISPs as much freedom as possible while ensuring that consumers do not lose out in the least on existing service, as well as upgrades to their service.

Thats in a world in which you accept the consumer as the person who really matters. The moment that you disagree with that, you're questioning the existing financial system...which is an altogether different issue/discussion.

I am very happy with the net being enhanced so that it can varry video traffic, say, much more effectively. The threat has been that enhanced capacity is sold such that one particular movie vendor buys the exclusive right to deliver high bandwidth services though my ISP. The means thae the movies I watch on the net are determined by cable provider (as now) and the deals they make with movie channels. If that bandwidth exists, I want it to support an open market for information and entertainment.

As for HTTP and TCP, they are end-end protocol - my computer and yours look at packets, but the routers in the net don't -- they only do IP. The changes you mention take place are oustide the Internet cloud, in the end computers. Inside the cloud, IP packets trundle across the net unknowing. The routers (in the original architecture) know IP, not TCP or HTTP.

Now the issue is whether the router will look at your packet and see it is a communciation destined for LandEnd and drop it because your ISP has a deal with LLBean.

OK, I think I see what your problem is. You say:

As for HTTP and TCP, they are end-end protocol - my computer and yours look at packets, but the routers in the net don't -- they only do IP. The changes you mention take place are oustide the Internet cloud, in the end computers. Inside the cloud, IP packets trundle across the net unknowing. The routers (in the original architecture) know IP, not TCP or HTTP.

In the 80s it was believed (by Cerf, et. al.) that network congestion control could be managed at the end-points, but practice has shown that's not the case. There's something called "Internet meltdown" that occurs. Here's part of RFC 2309:

The original fix for Internet meltdown was provided by Van Jacobson. Beginning in 1986, Jacobson developed the congestion avoidance mechanisms that are now required in TCP implementations [Jacobson88, HostReq89]. These mechanisms operate in the hosts to cause TCP connections to "back off" during congestion. We say that TCP flows are "responsive" to congestion signals (i.e., dropped packets) from the network. It is primarily these TCP congestion avoidance algorithms that prevent the congestion collapse of today's Internet.

However, that is not the end of the story. Considerable research has been done on Internet dynamics since 1988, and the Internet has grown. It has become clear that the TCP congestion avoidance mechanisms [RFC2001], while necessary and powerful, are not sufficient to provide good service in all circumstances. Basically, there is a limit to how much control can be accomplished from the edges of the network. Some mechanisms are needed in the routers to complement the endpoint congestion avoidance mechanisms.

The routers look inside TCP and don't discard ACKs.

Congestion increases latency, hence we need interior mechanisms - in the routers - to accomplish low-latency streaming. That's the main reason that "neutrality" in its strict form will never fly: we have to discriminate amongst packets based on the QoS characteristics to make these new applications fly.

The example you cite is actually illegal according to existing law and is not really germane to the issue that's before the US Congress at the moment. A more relevant example would be whether my ISP can lower the priority of voice packets streaming from eBay's Skype service to Best Effort while allowing its own voice service to use Expedited Voice.

That's a business issue, and applying the principle that you get what you pay for it's not particularly un-American to allow them to do that. And while there's certainly potential for abuse in this new world, it goes both ways: what if Google marked all their web traffic as Expedited Voice even when it wasn't? Under your view the ISP is powerless and has to honor it, and that's just not right.

Thank you for the link, I did read RFC 2309 (though dated 1989, not quite the 'original architecture'), and while section 3 only discusses queue management without mentioning peeking at TCP, and section 4 concludes that more research is necessary for congestion problems, I understand that congestion control in routers could well be best done giving preference to ACKs. It certainly seems intuitively reasonable, so I'll certainly accept your assertion that is now common practice.

But I think you misunderstand the concern I have about the legislation. I am not against innovation in the network protocols. I am not worried about favoring ACK packets, but certain company's packets. As I said above, I am happy that "We may pay for a higher or a lower quality of service. We may pay for a service which has the characteristics of being good for video, or quality audio." I am against a subtle commercially driven discrimination. You suggest: "A more relevant example would be whether my ISP can lower the priority of voice packets streaming from eBay's Skype service to Best Effort while allowing its own voice service to use Expedited Voice." Yes, here is an example of a connection between very close layers, the IP and the VOIP layers. As they are close, you can argue that they should be the same layer, and your ISP is in the business of voice just as much as data. (Actually, I don't like that argument, as I would prefer to be able to chose independently who I pay to connect through from voice service to existing plain old phone network. I'd also like to be able to use a service which uses a new standard encoding even if my local ISP hasn't adopted it yet).

But suppose you allow that, then, where do you stop? Suppose your ISP runs an online auction: is there any reason why it should support traffic to eBay at all, when it has its own auction service? Suppose it runs its own on-demand movies - why should it have to allow through HBO packets? Suppose it has its own search portal -- why should it give preference to Google's packets, when the customer has available its own search service? Suppose the degradation happens now, not only to Skype traffic, but traffic from video sources of stations with particular political views? What happens when your ISP's platinum partners establish favorable treatment for packets from sites with particular views on evolution? It is a slippery slope, and the bottom end is not nice at all. If there is a way of influencing the browsing choices of people, even slightly, there will be money in it, and when there is money in it there will be unscrupulous people trying to get that control. Do you really want to us to set off down that slope? Sometimes you don't know what you've got till it's gone.

OK, so we've narrowed the disagreement to one fundamental issue: can we allow ISPs to adjust packet (really "stream") priorities without opening the door to all sorts of abuse? In other words, how can we allow AT&T to reduce the latency of its own VoIP streams without doing harm to Google, Skype, Vonage, etc. (The same issue occurs with video, but the answer's the same so one example will do.)

I think the answer is to regulate broadband Internet access in the least intrusive way that prevents egregious abuse, and let the marketplace do the fine tuning. The COPE bill that the House Energy and Commerce Committee passed last week directed the FCC to continue enforcing regulations that protect these rights:

* To encourage broadband deployment and preserve and promote the open and interconnected nature of the public Internet, consumers are entitled to access the lawful Internet content of their choice.

* To encourage broadband deployment and preserve and promote the open and interconnected nature of the public Internet, consumers are entitled to run applications and use services of their choice, subject to the needs of law enforcement.

* To encourage broadband deployment and preserve and promote the open and interconnected nature of the public Internet, consumers are entitled to connect their choice of legal devices that do not harm the network.

* To encourage broadband deployment and preserve and promote the open and interconnected nature of the public Internet, consumers are entitled to competition among network providers, application and service providers, and content providers.

I think you would agree that these guidelines prevent a broadband access company from blocking access to lawful sites, so the blocking scenarios you describe are unlawful and not a worry. That leaves the "degradation" scenarios.

There are couple of reasons that "degradation" of service (meaning demoting streams from high-priority to normal, best-effort priority) cannot and should not be legally prohibited.

For one thing, it's absolutely necessary to police out-of-spec streams in order for high-priority, low-latency streams to work. The whole system of putting the short packets that VoIP uses at the head the internal queues has no value if normal web traffic, ftp, and Bit Torrent streams are also marked for high-priority. If everybody gets high-priority, nobody gets it.

And the other reason relates to bundling agreements that the ISPs want to offer their customers and that their customers want to buy. Is it criminal for the ISP to offer me best-effort service for $20/mo, high-priority for $40/mo that I can use anywhere, and a bundle of best effort with their own VoIP for, say, $30/mo? I think a lot of people would buy that service and it makes a lot of sense. It also means that the low-price users are getting a bit of a subsidy via the voice service, an objective good for the masses. But it certainly does put Skype at a disadvantage, but that's sort of inevitable given that Skype hasn't spent the money to build a network and AT&T has.

Skype then has a choice: they can continue hacking away and improving their performance like every other application on the net has always done, or they can make a deal with the ISP that pools the purchasing power of the their customers and enables them to get a $30/mo bundle through Skype. And then we have true competition again.

The big issue here is that the choices that need to be made between good practices and bad are very hard to make in legislation, which tends to be more like an ax than a scalpel. Anti-competitive practices are hard to identify until we have actual markets in which to measure them. So at this point it seems that the prudent thing is to ban only the most egregious abuses in law, and wait and see what really comes to pass as the new IMS networks are rolled out.

The good news is that the Telco networks enabled under COPE can only exist in jurisdictions already served by cable, so there will be no true monopoly in play here. And if the Telcos and Cable MSOs play rough, we can regulate them more and also build more Muni wireless, auction off more licensed spectrum, build BPL and WiMax, etc. At any rate we simply don't know enough about the management of priority-based streams on large, multiply-interconnected networks to impose lots of regulations in advance. So let's start the experiment and see where it takes us.

In its day the best-effort Internet was a radical experiment that violated a lot of good engineering practice about network efficiency, but it ended up succeeding because the need was great and the bandwidth was there. It's now time to take the next step in the evolution of the Internet's plumbing, taking it from free-rider status to truly the dominant network of the day.

That's where we are, trying to build an Internet that can pay its own way.

This is indeed where we disagree. You say "I think the answer is to regulate broadband Internet access in the least intrusive way that prevents egregious abuse, and let the marketplace do the fine tuning." I understand your point, but I believe that won't work, as if you degulate, you destroy the markets for IP and services over IP. The market for services on top of IP relies on non-discriminatory access to IP.

You say, "Is it criminal for the ISP to offer me best-effort service for $20/mo, high-priority for $40/mo that I can use anywhere, and a bundle of best effort with their own VoIP for, say, $30/mo? I think a lot of people would buy that service and it makes a lot of sense." Of course what is criminal depends on the law. But what is beneficial? Well, a lot of people bought the closed pre-internet AOL, Prodigy etc worlds before the open Internet. They got their connectivity and their information bundled for one low price. It "made sense". But it didn't hold a candle to being connected to the open web. And indeed, the market sorted it out.

However, if your ISP is free to set that $40 to any level it likes whenever the competition for voice services heats up, then there won't be competition for voice. Because it is a real pain to switch to a competing cable company, and in many towns you just can't.

I liked the idea of being able to change long distance carrier without getting a new phone line. I like the SIM card in my GSM phone which allows me to chose my telco and phones separately. Those things are good for me and for the industry, I think. When it comes to net, then the same applies, but there is another aspect. Network connectivity is access to information. Information is what I base my view of life and all my decisions on. It is the layer on which the market and politics and scientific layers of our society run. Control over a consumer's experinece is therefore a very powerful weapon. Hence the constant corporate and state (in some countries) attempts to get that control. When my ISP gives me that package, I have to use their phone service. What about their radio stations? Will they control also what news I listen to? Presumably. And the TV I watch? Presumably. That's not what I call the Internet.

TV is an incredibly powerful medium. (For example, the elections in the US seem to be largely a function of TV advertising). The total value of TV is currently low in that there is a small amount of information availoable from a small number of sources, even though it be delivered with irresistible effectiveness. The opening up of TV provision with IPTV could have a very refreshing effect on competition for TV. I'd be able to buy History channel without the ads, and listen to town meeting from any town, not just mine. This relies on the ability of the consumer to directly connect to any place. I think that is so important that yes, I think we should go to the trouble of separating the layers legally.

The reason that new things flourish on the web is that there is this open competition -- Google doesn't have to negotiate with ATT to start a new search engine. Now you are saying you don't want that openness for enhanced voice-capable net.

I wonder whether have taken a defensive position because you mistakenly felt that net neutrality was an argument aganst better high priority lower layer protocols. Now you know that it is not the case (here anyway), are you really suggesting that we drop the open competition for services above and below the IP layer for voice-capable net?

You say, "Skype then has a choice: they can continue hacking away [...], or they can make a deal with the ISP that pools the purchasing power of the their customers and enables them to get a $30/mo bundle through Skype." And then we have true competition again."

Ouch. True competiton? Competiton on the Web allows you to start small, write an app and run it. Your idea of 'competiton', where you have to negotiate with huge companies to start, is a sad one which would not have allowed the Web to happen at all, if it had applied to IP in general. Why do you think voice-capable IP doesn't need this open market to thrive?

What is Skype, in essence? (a) It is a computer program that uses the internet. (b) It is a way of finding the IP address of your friend and punching though his NAT box in this world where we don't all have domain names and IP addresses. This is where internet engineering is suboptimal. (I belive many more peer-peer applications would have sprouted were it not for this problem.) (c) It is a for-pay service for connecting through to the plain old telephone service.

When it comes to (a) getting software, I really would like to have the option of chosing this. When it comes to (b), the coordination service, that is a bug in the current internet, and there may be ways of fixing it which could be standardized so that different sytems could call each other at home. When it comes to (c) a VIOP-POTS gateway, that really can and should be a competive market.

Let the ISP offer me best-effort service for $20/mo, high-priority for $40/mo that I can use anywhere. If their independent business which sells (a), (b) and/or (c) wants to offer me $10/month as an incentive for using their service, because of the profit they will make through that service, then fine. But if they want to cut off other ways in which I run audio over IP, then no, that is not fine.

Why doesn't this fall under the right some might propose to strike any business deal and let the 'market sort it out'? Because of the difficulty and expense of cabling up a town, and the monopoly-like effect it has.

There are just two ways the service can be rolled out, like cable, or with the openness of the Internet. We already have cable.

Of course this internet will pay its way, because there is no way I can get any voice grade IP without paying for it, and so the market can set the price. And the market will work much better, as it will be independent of the services on top of it.

The whole VOIP market in fact will be able to take off. There will be market for teleconference bridge software and services with much more interesting features than now. There will be room for new genres of voice communication, say... streamed lectures with question and answer, ...stereo family meetings, things we haven't imagined yet. The end result will be a bigger total market and more profits for the ISPs, too.

I've taken a defensive position against "net neutrality" because most of what's been written in support of it is crap. One group of Net Utopians claims that the Internet is supposed to treat all bits and all packets as if they had equal value, and any examination of their contents or prioritizing as they pass through routers is a violation of the sacred principles of IP architecture brought down from the mountain by Vint Cerf and etched in stone as a holy covenant with The Almighty. David Weinberger and David Isenberg argue this line, as do their Cluetrain Manifesto buddies. Another version of neutrality maintains that prioritization is fine as long as the carriers open their high-priority queues to anyone and they don't offer higher-level services such as VoIP on their own. This is the general view of the Save the Internet Coalition. And then there's the Arianna view that neutrality has something to do with blocking web sites, and yours that it's to do with exclusive access to your net phone number. These arguments are so slippery I may as well be arguing for evolution against the Intelligent Design creationists; any time I win a debate I'm accused of attacking a strawman.

I want a replacement for IP that works like the ISO network layer protocols, one service for datagrams and another for reliable, connection-oriented service with hop-to-hop retransmission and reasonable multicast. IPv4 is an incredibly creaky protocol, poorly designed at the outset and clearly incapable of handling the needs of the next generation of applications. IPTV needs multicast service with real-time retransmission like we have in WiFi. The TCP/IP service isn't rich enough for this application and for the others you've mentioned. All it takes to make Internet multicast to collapse is a couple of dropped packets on one branch of the tree. A reasonable network would be capable of handling retransmissions internally and not have to bother the source each time they were needed.

Most of the complaining about neutrality is from application-level service companies like Google who would apparently be quite happy if the Internet stopped evolving as any change in the lower levels is a threat to their monopoly. I work at the transport layer and below, and with the kinds of applications the Internet can't support yet, like video conferencing. My field is caught in a stranglehold by Vint Cerf's peevish architecture.

You should also be aware that nobody's ever made money selling IP; the big companies that were setup to provide IP backbone services such as PSI and WorldCom have gone belly-up, and the only long-haul links you can get today are sold by voice network companies who are hoping for enough consolidation to bring prices to a reasonable rate.

There has to be profit in IP if the carriers on the last 100 miles are to improve their networks, and the networks have to be improved if we're to reap the benefits of real broadband. From a business standpoint the only way this can happen is if the carriers can offer services above IP that make use of unrealized capabilities in IP for Differentiated Services enabled by DSCP and complement IP with a more powerful layer two.

The Cerf architecture is missing a session layer, and that defect is responsible for a lot of the controversy today. The Telcos have a session layer, IMS, that enables a set of multimedia services that may be the cash machines they need to satisfy their responsibility to their shareholders while building out higher-bandwidth, low-latency networks. ISO had a session layer, but the Internet didn't need one since it was a free network for all in the Walled Garden of Academe of its first decade.

I'm very encouraged by your admission that IP has architectural flaws, and would like to know if you see any clever ways of addressing them.

And BTW, it AOL and Prodigy didn't compete with the Open Internet in their early days because there wasn't one; there was only a Closed Internet for researchers forbidden from using it for commercial purposes. The spirit of monasticism that era inspired is still with us, unfortunately.