[Update: AARGH. Mea culpa -- I had been working on some documents regarding the VIACOM -Google lawsuit, and then I started work on this posting about the VERIZON-Google announcement about "net neutrality," and so my initial posting erroneously talked all about the "Google-Viacom announcement," as many of the commenters pointed out ... My apologies for the confusion!]
I have been reluctant to jump into the fray here, because — well, because “net neutrality” is a really complicated subject, and because I’m on vacation, and I’d sure rather go for a bike ride, or lay down some more tunes, or head down to the pond, or . . . rather than sort through the sense and nonsense being written in light of the Google-Verizon announcement. But there is so much nonsense piling up out there, that I feel as though it’s worth adding a small note to the debate.
First of all, if you’re interested in the issue — and you should be, because important principles are at stake, and because the Internet is a public resource of immense value and unimaginable potential value, and because the debate over “net neutrality” implicates both these principles and the future of that public resource — and are trying to think through the implications of the recent Google-Verizon announcement, start here (the announcement itself) and then go here (Larry Downes’ excellent summary/discussion of what the proposal actually says, as opposed to hysterical pronouncements about what it might mean for the future of humankind). Although you would think, from much of the public discussion (see the articles in today’s New York Times), that Google and Verizon have cut a deal to chop up the Internet into small pieces and feed it into the fire, or at least that Verizon has agreed, for a fee, to disable all searches using any search engine other than Google, or that the two companies have cut some sort of a deal to give traffic from the other some sort of preferential treatment — the reality is otherwise; the proposal is simply an outline for a legislative/regulatory framework that attempts to jumpstart the contentious and largely moribund process that the FCC had already begun, months ago, and which has been mired in disagreement and controversy. And, again as Downes points out, the proposals themselves aren’t really that far away from (a) what the companies had already said in those FCC proceedings, or (b) the FCC’s own stated position(s) on the matter.
As to the substance of the proposal, here’s the underlying problem, in a nutshell. “Net neutrality” is a wonderful thing – at least if you’re careful to define what you mean by “net neutrality.” What I mean by net neutrality is equivalent to the concept of “end-to-end design” that has been part of the Internet’s architecture since its inception. I’m a believer — indeed, I’m a religious zealot — when it comes to e2e design; if you want to know the details, read my book. E2E means, simply, that the network itself does virtually nothing besides transporting bits from one place to another, leaving all processing to take place at the “ends.” E2E has “non-discrimination” built into it; the network can’t discriminate among the different packets of data (to put those originating from Bill Gates’ machine, or from Verizon premium subscribers, or from Google users, at the head of the transport queue, for instance) because to do that it would have to analyze the packet (to figure out where it came from), and that violates the E2E principle.
End-to-end is a principle of profound significance; it is one of the more important reasons that the TCP/IP network became “the Internet” — i.e., why this particular network grew as fast as it did and took over the world. E2E allows the network to focus on a single task — moving bits as quickly as possible. E2E allows anyone on the network to deploy an application that will run over the network; innovation, on an E2E network, can come from anywhere, without requiring any re-engineering of the network itself (because the network doesn’t care what the bits “mean” or what they contain, it’s just moving them from place to place). Losing the E2E inter-network would be, in my opinion, a catastrophic development.
So I’m all for “net neutrality.” The problem is that there are many things an E2E inter-network (like the one we have) can’t do that people want their inter-network to do and would pay to have it do, and businesses serving those people want to provide those things. Things like guaranteed delivery of packets; the E2E network can’t promise that your packet will arrive at its destination, because that would require the network to keep track of your transmission as it moves along, and that’s the sort of processing that E2E says the network shouldn’t do. Things like extra-high-speed transmission for certain packets (those with the “premium” label attached); the E2E network can’t do that, either, because that requires network processing (to determine whether a packet is a “premium” packet and should advance to the head of the queue). Things like a virus-free network — the E2E network can’t give you that because it doesn’t “scan” packets for viruses as it moves them along; you’ll have to worry about that for yourself (running an anti-virus application, for instance, at your “end”).
The problem then boils down to: is there a way to preserve the E2E network — the open, nondiscriminatory inter-network – while simultaneously allowing people to get the services they want? Now in fact, that’s not exactly the question, because we know the answer to that one. There are already thousands, hundreds and hundreds of thousands, of non-E2E networks that do lots and lots of internal processing and provide lots and lots of services the E2E Internet does not provide. Your cell phone provider’s network, for instance. Most corporate wide area networks, for instance. Obviously, if Verizon wants to build a separate network and offer all sorts of glorious services on it, it can do so. The real net neutrality problem is this: if Verizon uses the Internet’s infrastucture to provide those services, will that somehow degrade the performance of the E2E Internet or somehow jeopardize its existence? Put another way: if Verizon can figure out a way to provide additional services to some of its subscribers using the Internet infrastructure in a way that does not compromise the traffic over the E2E inter-network, why should we want to stop them from doing that?
There are, at the core here, some difficult questions of engineering, questions I’m not capable of answering or evaluating. If the answer is: “That can never be done; it is impossible to use the Internet’s pipes and routers and all the rest for any sort of “discriminatory” services without harming the E2E Internet” then I might be inclined to side with those who want to ban such services in the name of net neutrality. I’m not confident, though, that that is indeed the answer, in part because technological innovation can do things that I usually cannot foresee. A world in which the open, nondiscriminatory, E2E Internet is alive and well, and in which some users can obtain additional services that they want, is a better world than the one where such additional services are forbidden. The Google/Verizon proposal seems to take a fairly reasonable middle ground on this central question. Here are the relevant provisions:
“Non-Discrimination Requirement: In providing broadband Internet access service, a provider would be prohibited from engaging in undue discrimination against any lawful Internet content, application, or service in a manner that causes meaningful harm to competition or to users. Prioritization of Internet traffic would be presumed inconsistent with the non-discrimination standard, but the presumption could be rebutted. . . .
Additional Online Services: A provider that offers a broadband Internet access service complying with the above principles could offer any other additional or differentiated services. Such other services would have to be distinguishable in scope and purpose from broadband Internet access service, but could make use of or access Internet content, applications or services and could include traffic prioritization. The FCC would publish an annual report on the effect of these additional services, and immediately report if it finds at any time that these services threaten the meaningful availability of broadband Internet access services or have been devised or promoted in a manner designed to evade these consumer protections.
So it’s a presumption of harm, and FCC oversight to see the extent to which harm is occurring. I’m not at all sure that’s the optimal solution to the problem — but its not an outrageous attack on the viability of the Internet, or the death of the Internet as we know it, or anything else it’s being called (including by people who should know better).
Joe says:
David,
This is the second post where you referenced Google/Viacom (the copyright case) instead of Google/Verizon (Net Neutrality).
August 11, 2010, 1:56 pmChris Travers says:
Fascinating post. As someone who has some competence in network engineering, I will attempt to clarify your central question:
I don’t think this is an engineering problem but a definitional problem.
And so you are making a division which is remarkably difficult to define between a private internal network and the public internetwork. When we look at a corporate WAN, for example, the distinction here is not in technology or even end-to-end mapping, but it is that there is a physical distinction between these networks and the barrier between them is a router. When I say physical distinction, I mean that there are separate wires coming out of routers to the internet and to the corporate LAN.
Of course LAN’s and WAN’s are different. The LAN’s are connected together with essentially virtual wires extending across the internet between routers on each LAN. These routers, which also act to define the edge of the corporate firewall (which is a virtual concept, not a physical product), are thus both a part of the internet infrastructure and a part of the corporate network. It’s on these routers that intelligent services would generally be offered (premium queueing and the like).
So, if you are OK with that, then we have to ask how this applies to ISP’s and where the division is between an ISP’s network and the E2E internetwork. Part of the problem here depends on perspective. From the corporate LAN perspective, the ISP’s network IS part of the E2E network. But from the ISP’s perspective, they are merely offering transmission services to their customers, and connecting to either peering partners or upstream ISP’s.
I suppose one could make a distinction between peering and upstream but I am not sure that’s a technological distinction. We could also distinguish between customerinternetwork traffic and and between peering-partnerpeering-partner traffic, but again definitions break down.
The problem, as I see it, comes down to how you treat all these parties and where you see the E2E internetwork’s lines being drawn vis a vis the ISP’s own private network. The ISP, even a top-tier ISP which only offers services to other ISP’s is likely to see its own network as private, and feel like it should be able to use its own network to offer services to its customers as they want. The downstream ISPs are likely to see this top-tier ISP as being the E2E internetwork.
So I guess the problem is: is it possible to draw principled lines to define the E2E network? I have serious doubts that it is. I’d suggest instead a different direction, which would look at specific issues and delegate them accordingly. For example, discriminating based on endpoint network might be a problem. Maybe discriminating on data because it’s a virus payload wouldn’t. And so-on.
August 11, 2010, 2:03 pmanomdebus says:
Question for pro-network-neutrality folk.
August 11, 2010, 2:03 pmAre you also in favor of ipv6 as a means of removing the need for nat? Note, I am not saying nat should be made illegal, just optional.
mnarayan says:
While this is technically true, TCP does ensure delivery or notification of failure. I’m not really sure what the underlying network could add on top of that.
August 11, 2010, 2:12 pmLevi says:
Appreciate the post. An issue that went unaddressed though is why the Google and Verizon proposal advocates enforcement of net neutrality by the FCC only on wireline services and not wireless internet services. I would guess this is caused by Verizon’s desire to discriminate over its 3G service, but I don’t know. http://bit.ly/aq0jPR
August 11, 2010, 2:17 pmChris Travers says:
Look up “virtual circuits”
August 11, 2010, 2:21 pmEMB says:
So suppose Verizon offers its customers an internet connection which normally has only speed x but with the “additional service” of being able to watch streaming video content from certain providers (e.g. Google’s YouTube) at greater speed y.
This seems to fit perfectly with what they’re saying should be allowed, but on the other hand this is pretty much exactly what net-neutrality advocates are worried about (i.e. ISPs exploiting their oligopoly power to be able to make this sort of deal with content providers), just phrased somewhat differently (some traffic is an “additional service” instead of competing traffic being throttled).
The ISPs could then claim that customers are still getting a network-neutral “broadband” internet connection (just not one capable of delivering the same quality of video from competitors as from their content partners). While this behavior is clearly anti-competitive, from the passages you’ve quoted, it sounds like the restrictions on anti-competitive practices like this would apply only to the basic “broadband” internet connection and not to the additional services.
August 11, 2010, 2:28 pmmnarayan says:
Yes, TCP provides a virtual circuit. What’s your point?
August 11, 2010, 2:34 pmChris Travers says:
Ok. Let’s look at this slightly differently and see if the alternative is worse.
Cable TV comes to homes in what is essentially a converged network with at least three different components, maybe more. We could assume that there is a packet-switched data network component, an analog cable tv component, and a digital tv/video on demand componet. We could also add voice circuits with no real logical problem.
So…. Big Evil Cable TV Corp decides they want to have these deals with the content providers. So they do as follows:
1) Offer internet connection which upstream is barely capped by connection channel constraints, introducing latency and disturbing VOIP, video on demand, etc.
2) Offering separate virtual circuits for video on demand, voice, etc. on their cable lines. These are subject to upstream network availability, and have gateway services that connect to Netflix, Youtube, Skype, etc.
This strikes me as entirely network neutral and yet exactly the sort of thing that the proponents of network neutrality want to avoid. By moving premium services out of the E2E network and into virtual circuits on the actual physical layer (TCP/IP model terms, not OSI ones), then the same goal can be accomplished but without violating network neutrality.
So what’s gained?
August 11, 2010, 2:39 pmDilan Esper says:
Here’s my bottom line. As long as it actually is as the post phrased– offering additional services without degrading the speed and quality of the rest of the internet– I’d have no problem with differential service.
The problem is, the power to make some things faster also includes the power to make other things slower, and you probably need a very strong watchdog to make sure that the latter is not happening.
August 11, 2010, 2:39 pmmichlawcrimefighter says:
If a particular service is prioritized without impacting the existing channels of traffic, it is still discriminatory prioritization. The analogue here is adding a separate private toll road over Interstate 5. Yes it has no deleterious effect on the existing road infrastructure, but its construction still requires finite resources that could have been expended on the public road. The net neutrality debate is really a debate about how best to channel private investment into improving public infrastructure. Every dollar spent on building a second tier is a dollar not spent on the first tier.
Of course convergence of voice and tv with Internet already violates net neutrality in spirit. For instance, there is no doubt that ATT Uverse has a tremendous advantage (price point wise) over Skype. Skype must rely on your internet connection while Uverse can compartmentalize voice and internet over its fiber, a violation of net neutrality in spirit.
We can actually look to video services on the Internet as a historical case study. Youtube was revolutionary not because it delivered video on a website (that particular technology existed years prior), but because it was able to do so in a cost effective manner. Bandwidth was expensive, both for the content provider and for the end user. As private investment flowed into improving the underlying internet infrastructure and as broadband became more affordable, the cost of video over the internet fell.
We are currently in the middle of a second revolution in video, delivery of HD content over the internet. Both youtube and Hulu already offer select video in HD. In the next couple of years, HD video will become ubiquitous. Once again, we can look at improvement in internet infrastructure. Five years ago, Comcast’s maximum tier service was 6 Mbps. Now their basic package starts at 12 Mbps, and for marginally more, they offer double that.
Now suppose that incident to the Comcast, NBC merger, Comcast wants to deliver NBC content directly to the end user in 1080p (which roughly requires double the bandwidth). Now Comcast has two options. They can simply upgrade everyone’s connections to 24 Mbps. But by doing so, they are subsidizing public internet infrastructure. Hulu would also be able to deliver 1080p content and compete directly with Comcast. Alternatively, they can do what Uvese does with voice, they can channel it through on a separate layer for just Comcast customers. Why upgrade your internet speed at large and allow other content providers to compete, when you can directly channel the service to your end user?
For me then, the spirit of net neutrality requires both that the internet traffic itself remain unprioritized and internet content (defined broadly) remain on the internet backbone. The argument that adding additional infrastructure, as opposed to rationing existing infrastructure, benefits everyone, is spurious at best. If we believe in the benefits of net neutrality, particularly innovation, then we need to defend net neutrality in spirit as well as on paper.
August 11, 2010, 2:41 pmChris Travers says:
TCP does not provide a virtual circuit. TCP provides a virtual connection. It is still packet switched, no throughput guarantee, etc. If TCP provided virtual circuits, H.323 wouldn’t suck.
Virtual circuits, however, exist in network technologies such as ATM. (ATM would be run as a physical layer in a TCP/IP network, allowing voice circuits to co-exist with packet-switched networks with no problems.)
August 11, 2010, 2:42 pmChris Travers says:
So does that mean that Google Voice cannot be used to call ordinary telephones?
August 11, 2010, 2:48 pmruuffles says:
I’ll bet they won’t have a 250 gig cap on it like they do for the internet use. That is, hulu content will count towards the cap but comcast content will not.
August 11, 2010, 2:51 pmmnarayan says:
You appear to be confusing virtual circuits with something else (permanent virtual circuits perhaps?), but regardless, my question remains: what would the network provide in terms of guaranteed delivery beyond what TCP provides? (A PVC would actually appear to actually lower the delivery guarantee as packets could not be switched around an outage.)
Edit: I guess I should note that the creation of legal barriers to the provision of guaranteed throughput connections does strike me as a pretty large problem with what net neutrality advocates generally push for.
August 11, 2010, 2:56 pmmichlawcrimefighter says:
I am not totally familiar with how Google Voice so forgive me if I misconstrue it. As I understand it, Google Voice is really just a website interface for traditional phone service that Google is offering for free. Google Voice gives you a phone number. If someone calls that phone number, Google will then route that call to one of your existing phone numbers based on various settings that you can control on their website. Voice traffic is not actually sent over the internet. But even if it were at some point, that is to say it behaved more like Skype, that only furthers my point that services like Google Voice should be allowed to compete fairly with companies that control the fiber.
No doubt, net neutrality is a line drawing exercise. Voice and television create special problems because they were independent from the Internet until very recently. But the point I was trying to make is that we need to think about net neutrality in terms of its goals and not so much the technical, historical definitions, especially in the age of convergence.
August 11, 2010, 3:10 pmMike says:
I am curious, other than spinning, what the practical difference between offering full speed service to premium commercial content, while limiting speed to non-premium content (ie, what most net neutrality advocates are concerned about); and offering full service speed for normal content and extra bonus services for premium HD content…
It all sounds like semantics, and whether you are spinning it for or against ISP’s. Unless there is some specific technical reason that makes one view makes more sense than the other, I’m afraid I’m going to have to continue to regard ISP’s claims that net neutrality prevents them from offering premium content with a fair amount of skepticism. Especially since most actual policies I have heard discussed or implemented (such as caps on p2p traffic) are more the results of ISP’s over-selling under-developed networks with false promises of “unlimited” bandwidth.
August 11, 2010, 3:11 pmmichlawcrimefighter says:
Yes and Comcast can get away with that cap without mass customer revolt precisely because you are currently unlikely to reach that traffic limit doing legitimate things on the internet. My guess is most people who get close are downloading very large torrents on a regular basis and let’s face it, most of that is illegal copyrighted content.
August 11, 2010, 3:15 pmToby says:
A good balanced discussion, supporting commerce but recognizing that something is potentially lost….
An issue to me is trust. Today’s internet providers to the home have demonstrated again and again that they do not deliver what they promise. When called upon it, they always act as if they are the party with less technical sophistication and knowledge than Grandma.
What sort of download speed do you get at home? What sort of upload? I regularly get fast downloads d for something the size of a youtube video, but often not for other downloads. Uploads are slower (OK they do say asymmetric in the fine print), but why do they slow down when I do a big upload. Somehow, these speeds are now burst speed, and can’t be expected to post an entire technical PowerPoint, or upload an entire website, or restore a backup over the line. Oh no, that might look like file sharing, and so is locked down….
Then there is the continuous breaking of RFCs. RFCs are the essential rules of how the standards work. You may have seen ISPs breaking DNS to increase revenue. Instead of returning an error when a record is not found, they return success and route you to an advertising server. They claim this is a “service”, but it causes such actual RFC services as VPNs to fail. (If you do not get an “address not found error”, how will your computer know to fail over to the internal office DNS).
I am easily won over by arguments of freedom to innovate and competition? But when the people who are making them have demonstrated themselves to be dishonest, again, and again, and again, their arguments carry less weight with me.
And besides, we had several major environments based upon just the sort of walled gardens they now propose. Compuserve is essentially gone. AOL is essentially gone. I am reasonably sure there is not a consumer groundswell to bring them back.
August 11, 2010, 3:22 pmToby says:
Different topic on IPV6 and NAT. I have worked in rooms full of utility smart grid folks who see IPv6 as a means to gain direct control over everything in the house. I like that the limitations of IPv4 limit them. I am glad that the limitations of installing a NAT create some semblance of a security model, including security from the smart grid operators.
Law enforcement would like to have direct access to all operational data, using the precedent of cell phones. A truly flat network defaults the visibility of that operational data to full view. At least with a NAT, they have to be let in (like a vampire needs to be invited in the first time…)
Sure, IPV6 and routers and good security practices followed everywhere (especially by Grandma) could do the same…but NAT does have its benefits as well as problems.
August 11, 2010, 3:28 pmDan Weber says:
In theory this sounds good. How do we do it in practice? How do we tell if a network is “good enough”?
The ISPs could leave their existing networks alone and build new priority ones, but eventually demands on the existing networks will grow to where certain applications fail. Maybe you can’t make your VOIP call because I’m downloading a bunch of movies, unless you pay for the premium service.
Some net-neutrality folks have unreasonable expectations, that everything should always be the same, but shared networks quickly get crushed by the torrents. The torrentors will just say “well, make the whole network bigger,” but they will fill it up again.
You can’t really stop priority service. It already exists in some fashions, since websites using Akamai pay more to be delivered faster. You can set up your VPNs to go over the public internet for speed but have a backup leased line for availability.
I think the priority packets are inevitable, but with that comes many possibilities for abuse. I don’t think stopping this abuse is impossible, but it might be very hard.
August 11, 2010, 3:40 pmanomdebus says:
Thanks for the reply. I asked mainly because of David’s mention of E2E being a good reason for network neutrality. IPv6 seems to enhance that and single IPs for multiple machines degrades it.
IPv6 and NAT are fully compatible and even though you can have a static one to one relationship with public IPs and local IPs. you don’t need to. As it is, almost all commodity routers are shipped with common sense settings already applied. A new wireless router could be set out of the box in bridged mode instead of NAT, but the manufacturer decided NAT was the better choice. This will likely not change. Manufacturers that do try to change that model will probably have to deal with lawsuits when Grandma’s refrigerator becomes the largest torrent site around.
The benefit comes from being able to directly connect any server to one another no matter where each is, without convoluted stuff like UPnP.
August 11, 2010, 3:46 pmDan Weber says:
I think one top-down regulation that would get us pretty far is that customers can say how they want their traffic to be handled, and they can be billed on that. But they cannot be billed based on the peer, only on the type.
So I put my VoIP packets with a “send with best effort, drop if delayed too long” priority, and am billed accordingly. It does not matter to whom I am sending those packets, either TimeWarner or Skype or Vonage or my friend in Australia.
Gaming packets are probably handled the same way. Interactive shells want low-latency but can deal with occasional slowdowns, so I leave them handled as normal, pretty much like web pages.
Downloading an OS update is big and not very time-sensitive, so I get a discount for labelling those packets “low priority.”
I’m not sure about movies. My Wii doesn’t have very much of a buffer, but it has some, so I can handle lags of a few seconds here and there without noticing at all.
One problem this doesn’t deal with is that I can’t pay for a fast pipe someplace specific. If, say, YouTube or NetFlix has a data center 100 miles away, my ISP could build a trunk to them and now YouTube packets get handled a lot faster. This strikes me as a Good Thing, but the above commenters have pointed out how it can lead to Bad Things. Maybe YouTube should build multiple data centers and use a different network to speed things up to their various physical locations.
August 11, 2010, 3:56 pmChris Travers says:
How can they afford to give me $0.03 to Indonesia then? I’m pretty sure that intercontinental traffic is at least in some cases routed over the internet.
But the problem here becomes a question of gateways. If you say broadly define internet traffic, how do you separate out, say, voice data that’s eventually going to a TV and video data from Netflix that’s eventually going to a TV or similar?
Not really. I hate to do this because this is fairly technical but I guess I’ll go into what the problem is.
There are two basic ways of routing bitstreams between terminals, and both of these are used in networks which are completely ubiquitous to our lives. The first is packet switching (Internet) and the second is circuit switching (PSTN).
In a packet switched network (ideal for most data), data streams are broken up into chunks which are called “datagrams.” These datagrams are then encapsulated in metadata and made into something we call “packets.” These packets are then transmitted over the wire to the next router. The router determines from that metadata where the packet goes next and resends it. It’s slightly more complex than this but you get the idea. If the router is currently transmitting, it just queues up the packet and transmits when it can. Each packet is independently routed with no timeliness guarantee.
TCP adds something like this called “connections” on top of this, for error handling and the like. Basically there’s a handshake that goes on between the sides (itself via packets sent back and forth) and every time a packet is sent, the other terminal sends back an acknowledgement. Additionally the packets are sequenced so that the other terminal can re-assemble these in order. UDP (a “connectionless protocol”) just sends a stream of datagrams with the idea that if some are lost, the upper layers can figure out what to do.
Circuit switching is entirely different and this is where virtual circuits come in.
In the early days of the telephone network, circuit switching is exactly what it sounds like. Either a person or a machine (like a Stowger switch) would connect electrical conductors together and thus this would complete an electrical circuit which could be used to send analog voice data. To make this work, you’d either dial a phone number (activating the electromechanical machine) or talk to an operator.
When the telephone trunks were digitized, it became possible to route several of these “circuits” across the same wire. Phone switches do this by dividing the wire into timeslices and communicating with the switch on the other side where a timeslice goes. These virtual circuits could then be routed between phone companies while cutting down on the number of wires necessary. A single twisted pair, acting as a DS1 (aka T1) could carry 24 DS0 virtual circuits when before it could only carry 1.
The advantage of this system is that once the connection is allocated, it exists until it is torn down. While you are talking on the phone, you are GUARANTEED, absent infrastructure outages, for the bandwidth to be available for the duration. If that gets exhausted and the next person calls in, they get a message that says “all circuits are currently busy.” There is no graceful degradation as there is with packet switching, but voice doesn’t degrade gracefully in those cases anyway.
The same basic approach is used for digital cell services using either the old DAMPS system or GSM. CDMA is a bit more complex however.
So the question becomes what happens when you run converged networks. Typically there are two ways you can do this:
1) You can run a packet-switched network and make sure you have plenty of capacity, combined with traffic shaping (prioritized queuing), or
2) You can run a combined network where circuits and packets co-exist on the wire level (like ATM).
ATM is one example of the second approach. It is what’s called a “cell-switching” protocol and divides up a connection into timeslices called “cells.” Timeslices can be allocated for virtual circuits or they can be allocated for packets. Virtual circuits, however, once allocated remain allocated in each group of cells until the connection is torn down. Packets, OTOH, are just routed on a first-come-first-serve basis.
These aren’t esoteric terms, but they require a little bit of understanding of the nature of switched phone networks and how these differ from the internet, and what the challenges are in terms of convergance. The old OSI initiative was an attempt at a fully converged network and many of the protocols they developed (including H.323) assume the presence of virtual circuits. Mapping these to TCP connections poses a number of problems in terms of security and performance simply because they are not the same thing.
August 11, 2010, 4:00 pmMike says:
Ah, networking 101. I remember that class.
August 11, 2010, 4:04 pmHoward Gilbert says:
TCP provides end-to-end error recovery and retry. IP provides routing between the ends, but by itself does not provide error recovery. However, when a large number of IP packets are transmitted between two points over a fiber or other leased line, then optionally the carrier can choose to transmit these packets using a link-level protocol that does provide node to node error detection (CRC) and retry (HDLC usually). This way if someone in NYC is sending data to LA, then while that packet travels on a leased line between Chicago and Denver, if there is a glitch in the point to point link and the link has error recovery, then Denver will tell Chicago to resend the packet until it gets through. That way the TCP recovery between NY and LA is not required because individual link level errors are recovered at the link level.
While TCP can recover from a loss, there is considerably more latency managing the recovery end to end rather than at the intermediate node link level. Furthermore, once TCP senses a problem it may start to reduce your Transmit Window size which impacts performance and throughput. So sending IP data over a network path that provides guaranteed delivery of packets at the link level does provide better response and performance than using a network that discards damaged packets anywhere in the middle and requires TCP to do the error recovery. That said, in many cases an ISP or carrier will transmit packets node to node with local DLC error recovery simply because the link hardware provides the extra error recovery at no real cost.
One of the first discoveries about computer systems back in the days when performance was actually limited by what hardware could do was that systems should offer dynamic priority. High priority should be given to infrequent short units of data (interactive) and low priority should be given to continuous requests for high volume (background). Today every operating system from Windows to Linux applies some version of this to CPU dispatching priority for local processes and threads.
However, the TCP and IP protocols make it relatively complicated for intermediate switching equipment to take packet headers (source and destination addresses and port numbers) and translate them into high priority trivial requests (like messaging, mail, and simple web pages) and low priority bulk data transfer (big file downloads). As long as the internet continues to have more bandwidth in the middle than the aggregate demand of all customers, there is no need for user streams to be prioritized so the user gets first what he wants most, but if demand ever exceeds capacity this will become a requirement (as it has been for every aspect of computer processing for the last 50 years). Remember, this is about giving the end user the ability to request, or automatically receive, higher priority for the small amounts of data that he regards as most important over some other data he regards as less important just as he sends some packages Overnight Express while he sends other packages are using less expensive UPS Ground.
Comcast (and all other coax cable TV systems) provides around 135 channel frequencies, which in the modern digital world all carry 40 Megabits of digital data from the central office to every subscriber. It can offer Internet data service on 1 to 4 channels providing between 40 and 160 Megabits of data transfer capability. However, the other 131 channels are still used for broadcast and pay-per-view digital TV content transmission.
Obviously the HD digital TV transmission of the local NBC affiliate TV station (that you think of as Channel 4 or 30) receives real-time priority in its dedicated bandwidth on the cable frequencies compared to the same TV show transmitted from Hulu over your Internet connection. So in this example, simply by admitting that the primary function of a Cable TV company is to deliver Cable TV, you admit that some types of digital data transmitted by a commercial bandwidth provider as a differentiated non-TCPIP non-Internet service but over the same physical medium as Internet service will receive preferential treatment and higher priority than the same content transmitted over the Internet. Well this also applies to TV and other services offered by Verizon FIOS or AT&T U-verse. There is an Internet piece that you buy and a non-internet piece that the bandwidth vendor uses for other services that are charged separately. Neutrality may apply to the 6 Megabit piece of the bandwidth that you purchase as Internet service, but what the Vendor does with the other 20 or 100 Megabits (or the other 131 channels of 40 Megabits each) is not part of the Internet service you bought and therefore not subject to Network Neutrality rules.
The Internet was originally designed to be Inter-Net, that is to connect different networks belonging to different companies sometimes using different technologies. If I send data to my next door neighbor, and I use AT&T DSL, and she uses Comcast as an ISP, then my data will go to the AT&T network where routing protocol will locate the nearest point where AT&T connects directly to Comcast. Perhaps that is in New Jersey. Although we use Internet protocols and addresses, my data travels through the AT&T network of Internet and voice calls and dedicated corporate bandwidth, and leased TV connections until it is handed off to Comcast. At no point did this traffic ever traverse some device belonging to a separate entity called “The Internet”.
Internet routing chooses the best route for traffic. It may be that both AT&T and Comcast are connected to some higher level generic core network, but traffic between these two networks may run faster and more efficiently through a direct connection than through a third party.
Obviously traffic between my computer and a neighbor that also uses AT&T can be routed over a shorter distance and have better performance than traffic that has to go to New Jersey and connect between different network providers. This also means that Google or any other third party may gain advantages by transmitting data over its own leased bandwidth to equipment located relatively near me and then communicating directly to me over the AT&T network. You cannot and should not imagine prohibiting such advantages when large users and carriers make more efficient use of dedicated private resources.
So the Verizon and Google statement seems to be a reasonable step to define what can be regulated and what sorts of rules ought to apply. Since some companies seem to assume that every individual could utilize enough central bandwidth to watch any TV show she wants any time she wants across the country over “the Internet”, some sort of management and rules should be in place before end users learn that there is not and never will be enough bandwidth to support that business model.
August 11, 2010, 4:08 pmChris Travers says:
Just to note, phone networks are a perfect example of a smart network with dumb endpoints, while the internet is the reverse. However each network excels at some things and fails miserably at others.
August 11, 2010, 4:08 pmChris Travers says:
yeah. That’s pretty much where I am.
August 11, 2010, 4:19 pmHugh says:
Any regulations by the FCC would be a futile gesture, no matter what technical data they have obtained. This server is now the ultimate power on the internet. I suggest we use it.
August 11, 2010, 4:46 pmBill says:
with all due respect to lawyers attempting to engineer networks, I bet this reads to me an awful lot what software engineers writing about the law must read like to lawyers.
You are correct as to guaranteed vs. best-effort packet delivery, but the following is ridiculously wrong:
Please re-read “End to end arguments in system design” more carefully, particularly the “performance aspects” discussion.
http://web.mit.edu/Saltzer/www/publications/endtoend/endtoend.pdf
It does not argue that the network must be completely dumb with all smarts at the endpoint; rather, it argues that careful consideration of engineering tradeoffs must be made before deciding where to put functionality in the system as a whole, and that it does make sense to put *some* performance-enhancing features into the middle of the network when the benefit is high and the cost is low.
Some amount of priority queueing, traffic shaping, prioritization, etc., is entirely appropriate here from an engineering perspective *if the folks building the system want the network to provide that service*. Since the network as a whole is built by independent organizations acting together — sometimes cooperating, sometimes competing, usually doing a bit of both — we have a broad question here which is primarily political/financial issue, not technical. The technology does not dictate that this functionality must be excluded for a better network; rather, it’s one where a cost-benefit analysis must be done.
Ultimately, any additional complexity in the network must be paid for by the subscribers; if there is a real market for services, a more complex, more expensive network that doesn’t perform better than a cheap simple network will fail to be profitable.
If there is a place where the government could do a little work to create significant improvement, it is probably to discourage anticompetitive behavior (like certain popular cell phones which are artifically constrained to only work with a single wireless carrier).
August 11, 2010, 4:49 pmmnarayan says:
While it’s not a big deal, you appear to be using a fairly non-standard definition of “virtual circuit”; what you are referring to would more commonly be called “circuit switching”. The more common definition of “virtual circuit” is simply a protocol which allows higher level protocols to treat a connection as a circuit, even when the underlying physical transport may not be, hence the “virtual” in “virtual circuit”.
Of course I’m not going to claim to have done some sort of extensive survey on the use of these terms so YMMV.
August 11, 2010, 4:52 pmBill says:
Hah! Not even close!
Perhaps this is true within a well-run autonomous system (AS); but between AS’s, the choice of route is constrained by policy which is largely driven by who’s paying, and the inter-AS routing protocols such as BGP have much more to do with announcing reachability than in getting packets to take the globally optimal path.
Proper network engineering will mean that most traffic will take a decent path but we’re a long way from optimal routing in the greater internet.
August 11, 2010, 4:55 pmSoronel Haetir says:
I thought the domain advertising was the fault of the registrars. Regardless, yes, it’s a lousy practice.
August 11, 2010, 4:58 pmChris Travers says:
Even if you look at the Wikipedia page, it covers virtual circuits on the (OSI) Data link layer (i.e. TCP physical layer). The major advantage to low-level virtual circuits of this sort is that they can allocate bandwidth across the network at the router/switch level.
With TCP connections, you can’t do this. Consequently, with level 2 virtual circuits, you get first choice over bandwidth even before IP packets are routed.
August 11, 2010, 5:08 pmChris Travers says:
This is, unfortunately, one serious aspect of the network neutrality debate. There are many people who think that traffic shaping itself should be banned and are unwilling to even listen to engineering arguments to the contrary. After all, traffic shaping COULD be used to hurt competitors.
As I said before, it’s a definitional, rather than an engineering, problem. Don’t dump a legal problem in the lap of an engineer ;-)
August 11, 2010, 5:17 pmThe Ghost of Spaulding Smails' Booger says:
“meaningful harm” & “prioritization of Internet traffic”
I’m sorry, but these phrases are a secure foothold into disrupting the free flow of traffic.
Rather than spend hundreds of millions in legal fees defining “meaningful harm” and determining if the presumption is rebutted, they should focus on R&D and develop delivery system that has no impact at all.
August 11, 2010, 5:34 pmBill says:
there are a bunch of mechanisms out there to capitalize on typos:
0) applications installed locally on the client can catch this case and pop up a “helpful” page possibly including advertising.
1) register misspelled domains (“typosquatting”). Anyone can do this if you get there first. You may draw fire from trademark lawyers, though.
A loophole allowed this to be done very cheaply: http://en.wikipedia.org/wiki/Domain_tasting
2) Individual ISP’s, which typically handle DNS traffic through their own servers running recursive resolvers, can synthesize responses pointing at a landing page when they get a “name doesn’t exist” error. For example, Comcast does this but lets you opt out in a couple ways.
3) Verisign briefly inserted a DNS wildcard record to do the same thing as (2), before being told not to:
August 11, 2010, 5:37 pmhttp://en.wikipedia.org/wiki/Site_Finder
EMB says:
I had the impression that AT&T’s UVerse was actually a recent move in the other direction, though I may be misunderstanding what you’re saying.
In any case, if net neutrality is to be regulated, I think the regulation should focus on the sorts of deals that ISPs can make with other companies (with appropriate anti-trust action to separate their TV businesses from their ISP businesses).
August 11, 2010, 5:37 pmDan Weber says:
Maybe we need to get the people who sell pipes to not sell any other services; or to publicly declare those services so they can be monitored for anti-competitive behavior.
August 11, 2010, 5:55 pmChris Travers says:
I don’t know enough about the low-lwvel implementation of Uverse to say. The use of H.xxx protocols however doesn’t rule out low-level virtual circuits, nor does the use of DSL.
To give you a counter-example, I have FTTH via an ISP which provides me also voice and tv services. These are all provided on an ATM link which isolates these other services as virtual circuits. Consequently the voice and tv traffic is isolated from the internet traffic.
August 11, 2010, 6:13 pmBill says:
I don’t think most people want to have to buy a wireless phone from one vendor, a phone number from a second, voice mail service from a third, etc.; if you want to, you can unbundle but requiring it turns everyone into a system integrator.
If anything, the support costs for getting the pipe-only, bring your own phone, people set up would likely be higher than the costs for the “we supply everything” customers.
that said, I’m happily using a T-mobile “bring your own phone” plan right now.
August 11, 2010, 6:16 pmmnarayan says:
level 2 virtual circuit != virtual circuit ;)
August 11, 2010, 6:29 pmCathar says:
The reason this whole thing is even interesting is because it involves terrestrial Internet service to the home, which is a greedy yet nearly incontestable monopoly or duopoly nearly everywhere in the US, along with Internet service to mobile devices, which is a greedy yet nearly incontestable oligopoly everywhere in the US.
May people hoped wireless Internet service would provide enough competition to constrain the oft-demonstrated wickedness of the incumbent terrestrial mono-(duo-)polists. Google and Verizon are collaborating to dash those hopes by proposing that wireless Internet service should NOT compete with terrestrial Internet service. Besides leaving the (already justly hated) incumbents to continue to squeeze terrestrial Internet customers, the Google/Verizon proposal would make wireless Internet a testbed for anti-consumer carrier/content-provider “partnership” deals (e.g., Verizon delays packets to all search engines other than Google in return for a rakeoff from Google’s ad business). Once such deals have been road-tested on the wireless Internet the big players will try to deploy them on the terrestrial Internet.
I don’t find such dealmaking surprising. It was inevitable as soon as Google had enough market share that it could make stable deals with carriers that also have very large market shares. Switching costs for customers of both parties are high now and will go much higher if proposed carrier/content-provider partnerships go into effect. For example, if you switch your advertising from Google to Bing now you’ll both lose some eyeballs and save some money, but you can hope that Bing will gain market share and you with it. Once Verizon makes Google 3X faster than Bing for Droid users, Bing will likely never gain market share. You’ll have to advertise on Google to get any eyeballs and you can enjoy the bitter taste of paying for the bribes Google gives Verizon to keep Bing out.
Google’s opportunism isn’t surprising, but I don’t think it would matter if it weren’t government policy to sustain the various carriers’ oligopoly positions. For example, the government chose to auction wireless spectrum to the highest bidder. That maximized immediate government revenue (and private returns from attendant corruption), but at the cost of sustaining the oligopoly carriers– protected from competition by the government’s licensing scheme, they screw their subscribers to get the money with which to pay the government (and to pay themselves, of course). It could have been different, and at one time Google made modest noises suggesting a better arrangement: government should have raised money with taxes and let the public use the wireless spectrum. Alas, that would not have lined the pockets of all the people involved in the wireless auction scheme.
Terrestrial Internet competition is also discouraged by government.
Everyone knows that monopolists/oligopolists restrict output to sustain higher profits for themselves at the cost of a smaller overall market. That is exactly what the carriers and Google (and other big content-providers) have planned for the Internet. They plan to boost their own profits even if that means the Internet overall is smaller and less useful to people and especially if it means that new competitors will have trouble getting started. To big firms like Verizon or Google the prospect of new entrants is not a happy one. They don’t see the (presently unknown) “next Hulu/Facebook/whatever” as a glad prospect, they see it as unwanted competition. They would prefer to rule in Hell than serve in Heaven.
August 11, 2010, 6:48 pmJay Gischer says:
I think the core problem is that I don’t trust the likes of AT&T. I’m not too thrilled with Verizon, either. But it was the president of AT&T who said in his congressional testimony that Google was stealing their bandwidth.
That was game over for me. The door needs to be slammed on this guy, and his nose broken. So anything that sounds like it has any wiggle room for the “stealing my bandwidth” crowd is a non-starter for me.
Perhaps you’ve heard how in China, disfavored websites become very slow, or have spotty availability. I don’t want our internet to become a commercial version of The Great Firewall.
August 11, 2010, 6:55 pmChris Travers says:
????? Or are you just saying I wasn’t being specific enough?
August 11, 2010, 6:57 pmChris Travers says:
I think that’s the real problem. Network neutrality is not a substitute for competition.
August 11, 2010, 7:00 pmChris Travers says:
Too late.
I guess the internet detected the censorship and routed it around….
August 11, 2010, 7:01 pmRichard Nieporent says:
The way in which traffic flows over the Internet is more complicated than a simple IP datagram model. The ISPs use MPLS (Multiprotocol Label Switching) on the Internet core. MPLS is an overlay on top of the IP packet switch network that provides a virtual circuit like capability to pass traffic from an edge router across the Internet core to another edge router. MPLS provides traffic engineering by enabling the ISP to assign traffic flows to different virtual paths across the core. In other words MPLS enables the ISPs to provide different services to different uses by assigning their traffic to different flows.
There is a lot more to the Google/Verizon announcement than meets the eye. The assumption is that wireless traffic will be the dominant means by which users access the Internet. In other words 4G networks and smart phones will be the means by which the majority of people access the Internet. By excluding wireless traffic, they are removing what will be the bulk of the traffic from the net neutrality rules.
August 11, 2010, 7:33 pmHoward Gilbert says:
A recent development has been the “datacenter in a shipping container”. Microsoft and other vendors have packaged racks of servers and disks, plus cooling and power distribution, into a container you can put on a flat bed truck and deliver anywhere. Although intended to provide “cloud computing”, this also allows the distribution of data storage for companies like Google. An unresolved question is arrangements and pricing for co-locating such a unit in the parking lot of a main ISP switching center. The advantage is to put large server capacity where they can have low cost high bandwidth connections to the equipment that concentrates user communications.
Although this has the possibility of transforming the “information superhighway” into the “information trailer park”, distributing data closer to the end user has always been an efficient alternative and might have been more common had the Telecom bubble of the late ’90s not overbuilt the world with lots of still-dark fiber and capital costs wiped out by bankruptcy.
In practice, Network Neutrality may depend as much on establishing fair connection pricing from ISP switching center to anyone who wants to rent a concrete foundation and electric hookup at the Verizon Acres Network Trailer Park as it does to the current consumer-centric fixation on services over the last-mile wire to the house.
August 11, 2010, 8:05 pmToby says:
Registrars may or not provide DNS service. ISPs do.
From the barefuit home page:
The Barefruit Solution
Barefruit provides a comprehensive solution to error traffic, enhancing the user experience and generating revenue for ISPs. Uniquely able to resolve both DNS and HTTP errors, Barefruit provides highly relevant alternatives to the user, putting ISPs back in control of users’ address bars. Operating at network level, Barefruit technology enables ISPs to resolve error traffic across the customer base with no capital outlay.
Barefruit is the only error resolution service which uses patented* technology to analyse the page at the point of error, in order to determine highly relevant alternative sites for the user. Relevance lies at the core of the Barefruit solution.
Live with 15m users worldwide Barefruit provides a unique solution to both DNS and HTTP errors. Users no longer need suffer the frustration of ‘Error 404 Page Not Found’, effectively a dead end and a poor surfing experience. Barefruit creates value from otherwise worthless traffic, generating additional revenue for ISPs.
Personally, I hate the “helpful assumption” that dear luser is brain dead and uses only a browser for everything. A similar assumption lies hidden in many of the anti-neutrality efforts.
August 11, 2010, 8:39 pm1040 says:
i think it is not unreasonable and fairly simple to expect that when i ask for a 1 mbps internet connection to my home, that guarantee is met x% of the time when measured across y second intervals, and that i see a z% loss rate, when i am not bottlenecked at the server or close to the server i am downloading from. this is pretty much the implicit understanding of the bw deals sold by access networks today. i dont see why such an arrangement would prevent me from watching hulu, netflix etc at whatever quality my access link provides, or getting voip.
of course, isps can screw up their provisioning ensuring that guarantees are met only over timeframes of minutes, rather than 10s to 100s of ms (i.e. network timescales), and then claim that the only alternative is to buy some premium service. the only safeguards against this are strong regulation, at least until real competition emerges on the access side. the prospect of the latter depends on ubiquitous mobile data access, but we are far from that, and it is disturbing that this access is explicitly allowed to be non network neutral in the goog-vzw framework.
August 11, 2010, 10:54 pmChris Travers says:
Unfortunately this places things outside the immediate control of the network provider. Traffic use patterns shift and consequently one can expect capacity problems at least some times. So I don’t see a principled line here working, unless you want premium services over something like the PSTN and latency-insensitive services over the internet. But is that what people want? Is it tolerable to get these from the same people (say, Verizon, AT&T, or Comcast)?
August 11, 2010, 11:58 pmGuy says:
What? How do you figure out the baseline? The two things you’re differentiating between are just two characterizations of the same situation.
August 12, 2010, 12:45 am1040 says:
today’s access lines essentially provide some version of this. statistical multiplexing with the large numbers in typical access networks ensures that you get the kind of guarantees i was talking about. sure, there are capacity problems sometimes, which is why there is a x% of the time. otherwise, if they are selling 10 1 mbps lines, backed by an aggregated 1 mbps uplink, when they know that about 50% of the people are likely to be using 50% of their access b/w at any time, they are lying, plain and simple.
i have bought access connectivity for home, and for a server company and dealt with many isps on this. this is not complicated, and it is implicit in pretty much every b/w agreement end users sign today.
August 12, 2010, 12:49 amChris Travers says:
Ok, so you get full bandwidth x% of the time, but over what period of time? Shall we make the guarantee period, say, 1 year?
Also if a network discriminates against ALL packets by, say, adding a random delay to routing between 0.01 and 0.1 seconds on each packet, that would still render many services entirely unusable but still be entirely network neutral.
I suppose it would not affect email, surfing the web, or watching YouTube videos, but it would make VOIP entirely unusable.
A non-network neutral solution would be to offer a cheap connection guaranteed for 50% of the time to be usable at full capacity, and make you pay progressively more for more of a guarantee.
August 12, 2010, 1:07 am1040 says:
this is exactly what i said in my comment. you can make these restrictions meaningless, but there is a clear, implicit meaning for what a b/w guarantee means today.
actually a jitter of upto 100 ms per packet would make voip be pretty much the same quality as it is today.
in any case, the fact that isps can do shady things that are network neutral does not break the case for network neutrality. maybe what is needed is to make more explicit the guarantees that isps provide end users, and ensure that the codification meets the understanding we have today of what b/w means.
August 12, 2010, 1:22 amGoogle/Verizon and Net Neutrality « Philosophy On The Mesa says:
[...] of the network and allowing the network itself to do very little aside from dumb transport of bits. Here’s his take: The problem is that there are many things an E2E inter-network (like the one we have) can’t do [...]
August 12, 2010, 1:33 amEngineer says:
I love TCP, but it’s not really germane to the network neutrality discussion.
There are, at the core here, some difficult questions of engineering, questions I’m not capable of answering or evaluating. If the answer is: “That can never be done; it is impossible to use the Internet’s pipes and routers and all the rest for any sort of “discriminatory” services without harming the E2E Internet” then I might be inclined to side with those who want to ban such services in the name of net neutrality.
The “pure” e2e internet has already vanished ie.:
- providers that include VoIP as part of the service package are configuring customer devices to give higher priority to voice traffic (so that a voice call isn’t degraded if someone else is downloading
- the interconnects that service providers have with each other include “service level agreements” which specify that traffic in excess of certain levels is treated at lower priority
- most notoriously, providers are employing “deep packet inspection” to prevent peer-to-peer file sharing applications from sucking up all the capacity in the network
Technically it’s easy to use the same technologies to place a “high-quality toll lane” next to the “regular lane” of the internet infrastructure (this could amount to creation of two distinct MPLS “virtual circuits” in each provider’s network).
But providers could choose to give plenty of bandwidth to the “regular” traffic (as they do currently). Or they could starve the regular traffic of bandwidth, so that only the high-quality path offers the level of service we are used to. It’s not obvious to me which would happen, as it depends on market incentives, consumer acceptance etc.
The internet and web as we know them came about by accident, as the infrastructure was built for other reasons.
If not for this “accident”, what would have happened is that 80s vintage services like Prodigy, AOL, Compuserve etc. would have continued to build their proprietary networks.
Perhaps the “old guard” would have made more money, and consumers would not have the vast, cheap, and uncontrolled range of services that they have today. Or perhaps the proprietary online providers would have eventually collapsed into the free-for-all model.
Anyway, the network neutrality issue is not primarily technical, but rather trying to predict the unpredictable. It seems reasonable to allow market forces to try to innovate.
August 12, 2010, 2:45 am1040 says:
pretty sure the fcc ruled against this, and the providers resorted to the more transparent mechanism of absolute b/w caps.
this is a reasonable argument once access becomes competitive and changing between access networks is frictionless. we are very very far away from that world.
August 12, 2010, 2:51 amVerizon And Google Sitting In A Tree, Killing Net Neutrality « Around The Sphere says:
[...] David Post Possibly related posts: (automatically generated)Scour The Internets And Find Us A Creative, Funny Title About Net Neutralit…Neutral Tubes, Regulated Tubes, Free Tubes, Paying TubesStrange Bed Fellows: AT&T, Verizon join Google in net neutrality groupVerizon, Google partnership makes for strange bedfellows [...]
August 12, 2010, 9:20 amToby says:
1040 expresses my concerns more cogently than I was able to, and Chris T seems to be straining at gnats while agreeing. Even with peer to peer included, today’s providers are like the grocery store that advertizes free steaks with each jar of grilling sauce, and then begs out of it when they run out because they never thought anyone would hold them to it. They advertise throughput.
They offer it stochastically, which is reasonable. They never say, “Oh, except in the evening, when we plan to use all our bandwidth for Hulu” They never reduce charges when they did not anticiapte growth in a neighborhood. They are slow to subdivide neighborhoods into smaller realms after over-selling. They understaff customer service so timely complaints are impossible. The do this under the umbrella of a local government restricted market, and then they bray about competion and innovation. (See previous Volokh content with searches on FIOS and on Comcast)
For a good discussion of why it is their own position which is anti-innovation, I recommend reading yet another Law Prof, Jonathan Zitrain, especially his book “The Future of the Internet and How to Stop It”. The future he is trying to stop is the conversion of the internet into a 500 channel version of TV in the 50s, one way, no new products, no duisruptive technologies, and all of us consumers nestled nicely in our demographic niche.
For people who want to understand the pro-competio
August 12, 2010, 10:23 amChris Travers says:
This is where policy needs to be aimed IMHO. We don’t have enough competition in most of the country, and the policy needs to be pro-competition EVEN WHEN we do.
August 12, 2010, 11:38 amEngineer says:
The future he is trying to stop is the conversion of the internet into a 500 channel version of TV in the 50s, one way, no new products, no duisruptive technologies,…
I’m sympathethic to that view. But we generally fail to appreciate the strange combination of events that got use to where we are now.
Until 1988 the internet had no congestion control whatsoever and users were accidentally consuming all the bandwidth with retransmissions (the famous “congestion collapse”).
This just illustrates how the underlying technologies were developed without any sense of a business model, or of a vision of using it for mass deployment of high-bandwidth applications.
The truly amazing thing is that the underlying technologies (IP and TCP in particular) were so scalable that they could be adapted to the current global network.
But if the IETF people in the 80s had been business oriented in the slightest, there is no way anyone would build a network in which all packets are treated the same way.
Similarly, if they had known the future, the earliest ISPs would not have instituted the business model in which you pay according to the size of your connection rather than the amount of resources that you actually consume.
August 12, 2010, 11:39 amChris Travers says:
Ok, so here’s the problem in more concrete detail. Suppose a provider advertises a guarantee (or no monthly bill) of 80% of the time being able to meet capacity. Note this is the only metric I can think of that works– the time where the capacity can be met whether or not you are doing anything. That leaves an average of 4.8 hours per day when the upstream link is saturated as a maximum.
Now, suppose the ISP actually meets this goal exactly. Now, you may only be using your network connection 5 hours a day, and if that’s all during peak hours, you are going to be able to use Skype reliably during about 12 minutes on average of that time.
Suppose that goes up to 90%. Then half of the time, you’d be able to use Skype.
Now, if your phone is through Vonage things get really, really bad very quickly because this means that at 90% guarantee, you still have an average of 2.4 hours PER DAY where your phone is unusable.
If this is the concern one could come up with rules for advertising where congestion rates might need to be disclosed on a historical basis. This would provide incentive to keep lines free of congestion. However, I am not sure that’s enough as long as one only has 2-3 choices for internet service. Pro-competition policy needs to be a part of the mix too.
I’d also suggest that ISP’s should be required to disclose traffic shaping policies to their customers on demand. But again that doesn’t do much good here unless there is a clear move on the part of Congress to tell the FCC to foster competition there.
So I come back to the idea that there’s not a whole lot that can be done unless we start fostering competition. That needs to be the center of the policy. I’m lucky. I have a choice of at least 10 broadband ISP’s where I live. I think everyone else should be so lucky.
August 12, 2010, 11:55 amChris Travers says:
I’d add that we have to be careful in avoiding allowing too much discouragement for disruptive technologies. I actually think the disagreement is how to go about preventing the lock-down of the internet rather than whether to.
August 12, 2010, 12:11 pmEngineer says:
So I come back to the idea that there’s not a whole lot that can be done unless we start fostering competition.
Most countries require that the access provider and the ISP be distinct.
This helps a lot, because it’s much easier to set an ISP than to string copper or fiber.
August 12, 2010, 12:30 pmToby says:
The Carterphone voice recorder and the fax, the original disuptive technologies that gave us the internet. Carterphone case allowed connection of “unauthorized” equipment to the telephone system. Fax modems began un-planned-for digital transmissions over voice lines. Without them, we do not have the internet.
Most “No Net Neutrality” voices seem to be with those who argued against Carterphone and the Fax. Bell argued against them with almost the same words used today.
And Chris T ++1
August 12, 2010, 1:07 pmDilan Esper says:
What? How do you figure out the baseline? The two things you’re differentiating between are just two characterizations of the same situation.
They aren’t, but that’s why you need a regulator. You need someone to be able to say “this you can do because it doesn’t degrade service” / “you can’t do that because it does”.
August 12, 2010, 2:21 pmChris Travers says:
BTW, one simple engineering solution to David’s question would be to essentially have the ISP run a converged, twin network. One side of the network would be circuit-switched similar to the PSTN and the other would be packet switched, end-to-end (or reasonably behaving as such).
If you want premium services, you could have the option of running those services on dedicated (OSI L2)virtual circtiuts (for example, via CDMA, ATM, or the like). These virtual circuits could be then allocated and torn down dynamically, and when not in use, the bandwidth could be made available for packet transport via ATM, MPLS, and the like.
This is how the cell networks will do things anyway (via CDMA or GSM).
On the positive side, this leaves all network traffic able to choose between end-to-end modes (a more expensive, dedicated, guaranteed connection and a packet-switched it-gets-there-when-it-gets-there mode). It avoids the need to deal with traffic shaping, etc. And these decisions could be made on customer premises rather than at the network operator.
On the negative side, this ends the complete independence of the packet-switched network as such.
August 12, 2010, 3:22 pmEngineer says:
If you want premium services, you could have the option of running those services on dedicated (OSI L2)virtual circtiuts (for example, via CDMA, ATM, or the like).
I don’t know where providers could get suitable ATM equipment. Maybe Ebay?
On the positive side, this leaves all network traffic able to choose between end-to-end modes (a more expensive, dedicated, guaranteed connection and a packet-switched it-gets-there-when-it-gets-there mode).
Chunk by chunk even the PSTN is getting moved onto the packet network via pseudowires of various flavors. There’s certainly no point in putting VoIP back onto real TDM.
August 12, 2010, 5:04 pmChris Travers says:
I dunno. I know I have ATM connections as far as my house, and that the county has some pretty good ATM switches which they use to send this out to appropriate ISP’s. I know my TV and voice comes in over the fiber/ATM connection as discrete virtual circuits. I don’t know where they got the equipment however.
By extension I know that the four or five choices I have for phone service have ATM equipment that would suffice.
August 12, 2010, 5:28 pmHoward Gilbert says:
A DSL modem can run in “bridge” mode. It acts as a Layer 2 bridge transferring all Ethernet packets with non-local destination MAC addresses over the DSL line. Of course, for this to be useful the phone company end has to be configured to operate as a VLAN (Virtual Local Area Network). Typically you buy DSL as part of a commodity Internet Service, but at the hardware level it is Layer 2 divorced from IP and Internet service.
I cannot speak for modern Cable Modems, but back in the days before Ethernet took over, they used to create campus wide Local Area Networks out of Cable TV. That too was strictly Layer 2 stuff independent of IP. So there is nothing in any of the currently used network equipment that precludes the use of Quality of Service or timeslicing between the modem/bridge/router and the central service. If the vendor wants to sell me 1 Mb internet and use the rest of the bandwidth for something else, he is free to do so.
August 12, 2010, 6:30 pmMark Butler says:
May I suggest that one examine Title II of the Communications Act of 1934 (as amended) and existing case law about common carriers to get an idea of what Internet service providers are and are not going to be allowed to do?
Is there any reasonable possibility that ISPs are going to be prevented from doing in the near future what regulated common carriers are already allowed to do? Does anything about the Google – Verizon agreement except the suggestion that wireless traffic should be exempt violate the common carriage requirements of the latter?
I might add that the FCC brought this whole problem upon themselves by refusing to use their legally delegated authority in the first place, by initially classifying Internet service under a category for which they have no regulatory authority, letting the phone companies put thousands of independent ISPs out of business by eliminating shared access requirements, and instead creating a handful of local Internet monopolies riddled with conflicts of interest and each denying that they have any obligation to serve the public in a fair and non-discriminatory manner.
August 13, 2010, 2:57 amRichard Bennett says:
I hate to see people worshiping some abbreviated form of the End-to-End Arguments as if it were the sum of the Bill of Rights and the Law of Gravity.
Engineers don’t work from first principles, they work toward practical solutions to tangible problems, and the organization of functions in the Internet reflects this. End-to-End, which is some sort of post hoc theory about the organization of the Internet crafted by people who weren’t central players, is an argument for keeping systems general in order for them to be easily adapted to future applications.
It doesn’t have free reign over the entire discipline of network engineering, or even over the structure of research networks or other kinds of distributed systems. There’s also a principle that some answers need to be delivered within a finite time or not at all; this principle doesn’t have the power of a subset of the cyberlaw mafia behind it, but it should. There’s an E2E performance exception in any case.
The architecture of the Internet was cribbed from CYCLADES in any case, the research network built in France in 1972. E2E wasn’t a religious tenet of the CYCLADES designer, Louis Pouzin; he was just trying to build a nice sand box.
I have a problem with the rebuttable presumption against paid premium services in the GOOG-VZ proposal, and you legal wizards can set me straight on it. In my limited legal knowledge, rebuttable presumptions are more suited to ex ante adjudication than to the ex post procedure imposed by the GV proposal. Are we to assume that operators willing to provide premium service must build the system and offer the service before they can get a hearing regarding its legitimacy? Even for the kind of “mother may I” games the uber regulators want, that seems a tall order.
No wonder they gave mobile a full exemption.
August 16, 2010, 10:15 pmGeorge Ou says:
I don’t want to paste in a long reply, but I’ve posted an open letter for David Post to respond here.
August 17, 2010, 3:26 amhttp://www.digitalsociety.org/2010/08/can-we-drop-the-religious-zealotry-in-net-neutrality/
David Hess says:
NAT by itself provides no security over and above that of an IPv6 stateful firewall and provides no benefits except for conserving limited IPv4 address space poorly. Blocking incoming connections is a side effect of NAT created by the lack of state information for those incoming connections. If you want to duplicate the same effect with IPv6, you add a stateful firewall just like you would do with IPv4 if you did not use NAT. This is not a theoretical piece of equipment that is only available to networking professionals. Linux and BSD have had full IPv6 firewall support for years now and even Windows XP has been able to block incoming connections since Service Pack 1.
Besides the reliability problems with NAT which are about to get significantly worse when ISPs have to start using NAT for their customers because of lack of IPv4 addresses, the utility companies want a direct end to end connection without NAT so that equipment at the user end does not have to continuous poll the company end in order to maintain an incoming connection. The effect of enough deployed user equipment polling the utility just to satisfy the requirements of NAT would be the equivalent of a distributed denial of service attack. With IPv6, the user device contacts the utility company once (*) when it powers up and once every time its IPv6 autoconfiguration address changes if it takes advantage of IPv6 privacy.
Because of the way IPv6 autoconfiguration works, even without a firewall, searches by an outsider of someones local address space are not feasible unlike with IPv4. Any given device could be anywhere within the 64 bits of the local subnet. That is 4 billion times larger than the entire existing IPv4 internet and all located within one customers network.
If anything, law enforcement should be worried that adoption of IPv6 will lead to ubiquitous opportunistic authentication and encryption which would largely defeat CALEA in certain cases. ISPs could still forward intercepted communications but only the IP addresses and traffic volume would be legible.
(*) And as often as necessary for any IPv6 stateful firewall to remember to allow incoming connections. Oops. NAT would defeat the utility company’s desire for an end to end connection that does not require some type of keep-alive function but so would an IPv6 (or IPv4) stateful firewall. I have not seen anything in the trade magazines discussing this specific problem.
August 21, 2010, 11:28 am