My Epic Post On Biofuels
Okay, this isn't really about biofuels. It's about a major burr in my saddle: Net Neutrality.
Before I get into the technical, regulatory and economic meat, I must needs begin with a variety of analogies I've pulled out a number of times when people start acting all crazy about NN:
- When you order a supreme pizza, you get charged more than if you ask for a plain cheese. Pizza parlors often have other options such as ala carte toppings, individual slices and even salad! All available if you choose to avail yourself, all available for a price.
- Some people buy new Cadillacs with all the spiffy options (including crappy gas mileage), yet I see other cars on the road like 1985 Datsuns, 2007 Priuses, motorcyles, buses, and other vehicles. In many urban areas there are toll roads available if you want to go faster or more directly to certain surrounding areas that are not limited to Caddies, but for any car no matter how expensive or cheap or gas guzzling or biofueling it is.
- Say you go the Post Office to send a letter. You have a number of choices for delivery: if the letter is not time-sensitive and/or you're a cheapskate, you could just mail it First Class for 41 cents; should the letter be rather urgent, you might buy an Express Mail flat rate envelope for $16.25; add a larger item to the letter and it will cost you more because of the additional weight. First Class has always been "best effort" but generally arrives at some point, whereas the other higher-level options have guarantees associated with them.
As I tell my students all the time, every analogy breaks down at some point. People can't start comparing the specific realities of pizza making, for example, with telecom because of course they ain't exactly the same. Yet if you accept the basic premise that one may charge more for increased levels of service, then you're halfway there on Net Neutrality. If you don't accept that, well, that's beyond the scope of this discussion anyway--I'm not going to argue about basic assumptions of capitalism.
Now, I'm going to try to do this without the benefit of a week-long class that I teach on basic telecommunications, so some of the things will be grossly oversimplified (my colleagues and other bit-weenies better cut me some slack!). I hope the general concepts will be clear despite that.
So what the heck is NN, anyway? I find most people I argue with have a vague notion about it, not unlike the old line about porn, I know it when I see it. Fortunately Wikipedia has a good breakdown of the three basic flavors:
Absolute Non-Discrimination: Columbia Law School professor Tim Wu: "Network neutrality is best defined as a network design principle. The idea is that a maximally useful public information network aspires to treat all content, sites, and platforms equally."[2]
Cardozo Law School professor Susan Crawford: insists that a neutral Internet must forward packets on a first-come, first served basis, without regard for Quality of Service considerations.
Limited Discrimination without QoS Tiering: American lawmakers have introduced bills that would allow Quality of Service discrimination as long as no special fee is charged for higher-quality service.[10]
Limited Discrimination and Tiering: this approach allows higher fees for QoS as long as there is no exclusivity in service contracts: If I pay to connect to the net with a given quality of service, and you pay to connect to the net with the same or higher quality of service, then you and I can communicate across the net, with that quality of service.[1]. [We] each pay to connect to the Net, but no one can pay for exclusive access to me.[11]"See Sir Tim Berners-Lee.
I'll address these in a bit. First, I need to talk about what QoS is since it's featured so prominently in each definition. An explanation from my old company's wiki:
Quality of service (QoS) is a network's ability to support varying levels of network performance that can then be mapped to the needs of the applications supported by that network. The performance parameters we seek to control include such things as delay across the network, variations in delay, and total bandwidth available for a connection or information flow, to name a few.
Basically what we're trying to do is use a single networking infrastructure to deliver different applications that have rather different requirements in terms of service from that network. For example, voice hates delay. Imagine such a conversation:
Person 1: I love you.
PAUSE
Person 2: I love you, too.
That delay imparts meaning in a usual voice interaction--were you not listening to me, did you have to think before saying you love me, etc? Thus we want a network to deliver our voices without adding any delay (nor delay variation) because that could in effect change what we are trying to communicate.
On the flip side, consider an e-mail that says, "Come get bagels on the 3rd floor." Whether that takes 10 seconds or 2 minutes to deliver, the message will always retain its intended meaning. That's not to say if there's a lot of delay the bagels will still be there after your officemates devour them, but that ain't the issue. We're simply talking about the needs of e-mail and voice in terms of retaining their meaning.
So each of those applications have certain expectations and requirements of the service provided by a network. Voice wants low delay, e-mail doesn't care about delay. They also have different needs regarding other aspects of the network, such as errors--voice is rather tolerant of errors (the human ear can figure out meaning even if parts of the "message" are lost), whereas data needs to be 100% correct (it would be bad if you sent a note promising to transfer 100 bucks and it became 1000).
In the past, we've had separate networks for different applications. There was a voice network, for example, and a variety of different data networks basically tuned to the requirements of each specific type of application, and ne'er did they meet. Yet now we're using the public Internet and Internet-like private networks to run multiple applications together.
What providers are doing in the enterprise space is deploying IP networks with some extra capabilities that are not inherent to IP. The Internet Protocol is "best effort" and provides absolutely no guarantees of delivery across the network--obviously stuff gets to the other side most of the time, otherwise the Internet wouldn't work at all, but other mechanisms (the TCP in TCP/IP being a major one) on the ends deal with that. That's fine when pretty much all applications are cool with being jumbled together and have essentially the same needs for delivery, which is what the Internet environment was all about decades ago when IP was first implemented.
Nobody thought about running voice or video over the Internet back in the Dark Ages (you know, the 80s). Text reigned supreme and nobody was really running very complicated stuff. Today we are continuously coming up with new applications (i.e., uses) and they diverge quite a lot from the old, quiet, tradition-bound world of the pre-web world.
So we need new approaches within the network itself and that's just what the providers are implementing. In the enterprise environment they sell the same old vanilla flavor "best effort" service (marketed with better-sound names life "basic service" or "standard service") and for a little more coin, corporate customers can buy a "premium service" that offers differentiation between data from their various applications so they can improve overall performance and ultimately, the end user's experience.
An example of what happens if you treat everything the same in your network: Champlain College implemented a Cisco Voice-over-IP solution a couple years back for the usual reasons, including cost savings. A major, unanticipated issue was a severe drop in voice quality at certain times of day--what some people called "the monster on the other end of the phone" problem. They'd try to make a call, but for some reason the person on the other end would sound horrible, almost scary.
Turns out that students were getting done with their classes and firing up programs like BitTorrent and their interactive gaming stuff and effectively bogging down the network so much that the little teensy packets filled with voice samples were getting overwhelmed. Lose too many of those things (most such networks just throw out packets if there are too many), or have them caught behind a lot of big data packets (think of waiting at a RR crossing for a long train) and you can't reconstruct the actual voice on the other phone.
Solution? Start giving voice packets some logical differentiation on the same physical network. They examined the traffic patterns, tweaked things in their network devices and POOF! The Monster was exorcised and the students never noticed any change in their gaming or perfectly-legal (ahem) downloads.
So now I can address one misconception about how the Internet currently works and will work in this Fantabulous New Age of Quality. One thing I hear from folks is they're afraid AT&T will create two Internets: a highspeed King's Highway one for rich people, and a lowspeed, crappy, pothole-pitted Low Road for us peasants. In many minds The Evil Empire is literally building separate physical networks and will "discriminate" between who can pay for access to the Internet and those who will be shuffled off onto the back roads where you can't get to your favorite political blog and so forth.
It's actually a single infrastructure that AT&T has deployed to carry everybody's stuff, and it's just a matter of what CONTENT PROVIDERS, not end customers like you and me, pay for in terms of levels of service from that one network. So YouTube can stay with the traditional "best effort", which means no additional cost to them and we can continue to enjoy embedding fun videos in our blog posts. Yet if they wanted to start selling a premium high-definition movie service over the same Internet, they could put some extra skin into the game and have their video perhaps delivered in such a way that it looks lovely when you watch it on your computer.
I think this is where people get off the bus because they don't see how this can happen without "degradation" of our regular service. Perhaps a simple drawing will help:
Let's say I've got a router in my IP network that is receiving three types of traffic, each at different rates and with different needs. One of the things that happens inside the router is it looks at the "class" each packet belongs to and then puts them in appropriate queues. Depending on the rules the router's configured with, it will then pump these packets out in a "fair" fashion based on how much stuff is in each queue, what the QoS parameters are that each application has contracted for, and choose the appropriate paths for those packets.
So maybe the HD video gets put onto a path that offers high bandwidth and the voice and e-mail are interleaved on a lower speed path that still meets their basic requirements. Possibly some of those green packets will be shunted off on the lower path, too. Yet each colored packet will get access to the rest of the network, albeit in an arbitrated fashion to make sure the green video stuff doesn't take up all the room in the "tubes" (thanks, Senator Stevens!).
Voice and video, being more sensitive to delay, might get a bit more space than the e-mail, but the e-mail will still be delivered fairly. Yes, perhaps it will get slightly more delayed than if this were solely a "first come, first serve" algorithm, but we're talking milliseconds here, which are far below the threshold of any human to notice. All the while this is improving the experience for people watching video or using Skype or Vonage to talk to friends in Oz.
Now it would certainly be reasonable to expect if YouTube wanted to offer a new, high-quality video service that they would charge more for the increased value. That better, premium service is going to require extra quality and priority delivery across the network, which has cost the network providers money to implement with new, fairly pricey equipment. So why shouldn't some of what they charge us go to the providers upon whom they rely? Honestly, I still find it a puzzlement that people raise such objections, but perhaps that's because there are some other misconceptions about the Internet of which I need to disabuse them.
First, the biggest worry people seem have is that somehow AT&T is going to block political blogs, or create such conditions that it will "take 10 minutes to load Eschaton," thereby limited political speech. I hope at least from the QoS discussion you can see that there's no technical reason why that would be the case. As for the potential to begin filtering content the way countries like China and Vietnam do, well...the additional features providers want to charge for aren't necessary if they wanted to play along with a fascist regime, nor would the Net Neutrality language I've seen address the issue.
It's trivial to block sites, specific content, etc. Yet we haven't seen any of that without Net Neutrality because, as it turns out, Net Neutrality in its purest form already exists: no provider wants to be in the business of filtering content and, in fact, aren't, because that would destroy the very thing they want to make money from, namely the Internet. Anytime a provider has done bad things, such as blocking VoIP phone calls from a competitor, the FCC slapped them down with current regulations. That won't change.
Second, some people have expressed concern that AT&T will eventually stop supporting the old free services in favor of only the premium pay services. I suppose at some point one can expect all older services to fade away as more people move to the newer, splashier offerings. But the level of penetration in the market is low in the beginning of any product lifecycle (Fischer-Pry, anyone?), and as more people adopt the new technology, it will become cheaper (as we've seen with residential broadband deployment).
Beyond that, however, there's still a technical reason "best effort" won't go away: there's no benefit to the provider to get rid of it. IP inherently supports it, so there's really no way to turn it off. Remember, we're not talking about separate infrastructure (in contrast to AT&T's old TDMA network versus the newer, better CDMA network that Cingular had).
Third, somehow people think that AT&T is trying to make money off something that should be free because "the government built the Internet with our tax dollars." Nope, the government hasn't been in the networking business in years, and even when they were (e.g., with the NSFNet quasi "backbone"), providers were still building their own infrastructure with their own capital without government assistance.
Yes, the DoD did originally fund ARPAnet for research purposes many aeons ago, and out of that research came the TCP/IP suite of protocols that we use today. But TCP/IP is still free and is running over networks that the DoD had nothing to do with. The Internet is really not one monolithic network, but really a "network of networks" where AT&T's network connects to Verizon's and Deutsche Telecom's and even Ma And Pa Kettle's Basement Internet Service. They all use the same protocols so you can send me e-mail to bitch about this post even though my ISP is different from yours (not to mention I'm on dialup while you've got DSL, you lucky bastards).
One of the best examples of knee-jerk reactions against the providers is when people were freaking out in the wake of the patent infringement decision in a suit Verizon brought against Vonage. Folks who don't understand patents were scared that somehow Verizon was claiming a patent "on the Internet". Nothing is further from the truth, but because people naturally (and with great cause) distrust corporations and don't understand some fundamentals, they assume somebody's trying to pull a fast one. They ain't. At least, you know, where the Internet's concerned.
Anyway, many bloggers have been decrying the recent FTC decision regarding NN:
In the absence of significant market failure or demonstrated consumer harm, policy makers should be particularly hesitant to enact new regulation in this area.
That's essentially in the same camp as the FCC:
Kevin Martin, the FCC chairman, said during his portion of the keynote address that he believes the FCC's existing principles are sufficient to address problems that may arise should network operators block traffic.
"I think the FCC has authority to act," he said. "And it has done so in the past."
That's where I sit because having Congress try to legislate anything regarding fast-changing technology is a recipe for disaster. Really, do you want Ted Stevens authoring laws about "the tubes"? Nope, neither do I. And as the FCC and FTC already can deal with any truly unfair practices by providers, I'd rather follow the "first, do no harm" guidance when it comes to new regulation.
So back to those NN definitions. First:
Absolute Non-Discrimination
This is completely unreasonable and impractical in a modern Internet that is supposed to deliver new, different innovative applications. Hopefully the stuff you know about QoS will make that conclusion fairly obvious.
Limited Discrimination without QoS Tiering
Why do Democrats hate capitalism?
Seriously, here's one example of some the zombie legislation wandering the halls of Congress:
With respect to any broadband service offered to the public, each broadband service provider shallâ
(1) not block, interfere with, discriminate against, impair, or degrade the ability of any person to use a broadband service to access, use, send, post, receive, or offer any lawful content, application, or service made available via the Internet;
(2) not prevent or obstruct a user from attaching or using any device to the network of such broadband service provider, only if such device does not physically damage or substantially degrade the use of such network by other subscribers;
(3) provide and make available to each user information about such userâs access to the Internet, and the speed, nature, and limitations of such userâs broadband service;
(4) enable any content, application, or service made available via the Internet to be offered, provided, or posted on a basis thatâ(A) is reasonable and nondiscriminatory, including with respect to quality of service, access, speed, and bandwidth;
(B) is at least equivalent to the access, speed, quality of service, and bandwidth that such broadband service provider offers to affiliated content, applications, or services made available via the public Internet into the network of such broadband service provider; and
(C) does not impose a charge on the basis of the type of content, applications, or services made available via the Internet into the network of such broadband service provider;(5) only prioritize content, applications, or services accessed by a user that is made available via the Internet within the network of such broadband service provider based on the type of content, applications, or services and the level of service purchased by the user, without charge for such prioritization...
Part 1 I'm cool with, though the "lawful" content can open up a can of worms. Part 2 is also fine, though it's probably unnecessary in light of Carterfone. Part 3 makes me scratch my head, though it's certainly not fundamentally offensive.
But all those sections seem to be much ado about nothing and are not what NN advocates are up in arms about anyway. Parts 4 (specifically section C) and 5 are what I take issue with because of the charging issue. Which brings me to the last option that Sir Tim advocates:
Limited Discrimination and Tiering
This is coming from somebody who understands the Internet from a technical perspective, as well as the implications of new applications on the network. It's a balanced approach and I think when people get some of the fundamentals, it's the one that everybody can find common ground with.
The funny thing is that ultimately being able to charge for differentiated services ain't gonna do providers much good anyway. The real money is in developing applications, not delivery per se. But they do need to recoup investment in the network now to bridge them to the application-centric era in which they will be competing.
Long post boiled down: decisions by governing agencies to reject adding Net Neutrality regulations do not mean "we're fucked," as I've seen some in the blogosphere breathlessly claim. It's reasonable to avoid imposing anything new and unnecessary as we continue to develop a modern networking infrastructure (not unlike the moratorium against Internet taxation that was in force for so long).
ntodd
PS--As with all my long posts, I will be fixing the inevitable stupid things and also be updating in response to queries and new thoughts.
[Update: NYMary was asking me a few clarifying questions, so let me add one more boiled-down note.
The networks are NOT looking at the content in the IP packets. They don't care what the content providers are sending. The content providers, however, are marking their packets with the priority levels they want and then the network delivers that data according to contracted rules. So the decision about prioritization comes from THE CUSTOMER, not the network providers.]
I was going to say this the other day at that other website, but decided to stay out of it. The uproar over this makes no sense. What would be the benefit to companies to make the internet worse?
The internet in your home/office/business isn't free. A lot of money is being made with online services, why would a provider price customers out of the market when consumers still have options?
And you mentioned dialup, not everyone can get DSL, or wants to pay for cable, so there is already a tiered system of quality in place.
Posted by: silver | July 07, 2007 at 07:06 PM
I posted about this over at Candleblog.
Posted by: bill simmon | July 08, 2007 at 04:56 AM
Love the post. One suggestion for the diagram: show the serial flow of packets one at a time out the eventual tube:
G B G R G B G B
That makes it clear that all the packets got through in a mostly-green-but-still-plenty-blue (and look, the entire red got through pretty early) fashion.
(You may even want to simplify further and take the two channels out -- those are just algorithmic anyway, and real QoS algorithms are already way more complicated than simple queues.)
Posted by: Alex | July 08, 2007 at 09:17 AM
Thanks for the patient and rational analysis. You've allayed the fears I had about threats to "net neutrality". Now if we could only get the NSA off our backs. [;-)]
Posted by: Winooski | July 08, 2007 at 10:24 PM
I think the biggest reason people are upset about proposals to prioritize content (and charge for that prioritization) is that it upsets current business models, i.e. Google is upset that their costs will go up and end-users are afraid that content providers will have to charge for things that are currently free.
That aside, I'm not sure I understand why you think this won't be a problem (why you think that Absolute Non-discrimination is "completely... impractical in a modern Internet that is supposed to deliver new, different innovative applications," yet you're sure that non-prioritized content will receive enough CPU, NIC and network bandwidth to have their "requirements met"). I mean, fundamentally, isn't the issue that there are limits to 1) the number of CPU cycles available; 2) the size of network inteface buffers available, and 3) the # of packets that the network medium can transmit error-free? And if you're prioritizing some content such that they get more of those, how do you know that other content will get "enough." As an example, say a small political website wants to transmit video but hasn't paid to have their content prioritized. Their video streams therefore suck at times when Netflix is transmitting lots of HD video streams, right?
Finally, why do you think that it won't be necessary to ensure via regulation that the providers "enable any content, application, or service made available via the Internet to be offered, provided, or posted on a basis that .. B) is at least equivalent to the access, speed, quality of service, and bandwidth that such broadband service provider offers to affiliated content, applications, or services?"
Posted by: smoothie | July 09, 2007 at 04:57 PM
That aside, I'm not sure I understand why you think this won't be a problem (why you think that Absolute Non-discrimination is "completely... impractical in a modern Internet that is supposed to deliver new, different innovative applications," yet you're sure that non-prioritized content will receive enough CPU, NIC and network bandwidth to have their "requirements met").
Because the routers can switch at wireline speeds and because QoS is sophisticated. It won't be a problem because of the nature of network design and traffic engineering.
Why do you think that it won't be necessary to ensure via regulation that the providers "enable any content, application, or service made available via the Internet to be offered, provided, or posted..."
Such regulation already exists, which is why the FCC and FTC have ruled there is no need for further legislation or regulatory rule changes.
If such a thing is required in the future, then deal with it. Otherwise, this is a solution in search of a problem.
Posted by: NTodd | July 09, 2007 at 06:57 PM
Alex - good points. I originally had two drawings and was going on a slightly different tack, but ended up changing things a bit and didn't feel like editing the graphic. Still don't, but your comment should elucidate enough. Thanks!
Posted by: NTodd | July 09, 2007 at 09:49 PM
Because the routers can switch at wireline speeds and because QoS is sophisticated. It won't be a problem because of the nature of network design and traffic engineering.
That's just marketing speak and amounts to "trust me." While it's comforting to know that QoS is sophisticated, I was actually hoping for a more sophisticated explanation that didn't ask me to have faith in magic technology wielded by benevolent corporations.
For example, you say that the routers operate at wire speed. All that means is that the router can switch packets as fast as the network on either side can transmit them. Are you asserting that all these routers do is rearrange packet order to smooth out bursts of non-prioritized traffic and never drop packets or send ICMP source quench messages? Seems unlikely. And if they do drop proportionally more non-prioritized packets or tell non-prioritized senders to back off, then that is degradation of service and my question is, on what basis do you assert that that degraded service is "good enough?"
Posted by: smoothie | July 10, 2007 at 12:02 AM
Such regulation already exists, which is why the FCC and FTC have ruled there is no need for further legislation or regulatory rule changes.
Please pardon my skepticism that any Federal regulatory body composed of people appointed by our current president is adequately protecting the public trust.
If such a thing is required in the future, then deal with it. Otherwise, this is a solution in search of a problem.
"An ounce of prevention is worth a pound of cure." Ask Netscape's shareholders how they like their timely remedy in the MS antitrust enforcement.
Posted by: smoothie | July 10, 2007 at 12:17 AM
That's just marketing speak and amounts to "trust me."
No, it's not. I've explained QoS technically, thanks.
While it's comforting to know that QoS is sophisticated, I was actually hoping for a more sophisticated explanation that didn't ask me to have faith in magic technology wielded by benevolent corporations.
What more sophisticated explanation do you require about QoS? Do you really want to delve into queuing theory right here? Should I go into DiffServ marking and how that will impact CBWFQ at the ingress? Are there other fancy terms you'd like me to use, and would you accept them or simply reject them as well?
For example, you say that the routers operate at wire speed. All that means is that the router can switch packets as fast as the network on either side can transmit them.
Yes. Thus, they can crank out packets at extremely high speeds on individual ports. Backbones are operating at OC-192 on single lambdas currently. Go ahead and try to choke that.
Are you asserting that all these routers do is rearrange packet order to smooth out bursts of non-prioritized traffic and never drop packets or send ICMP source quench messages? Seems unlikely.
Then perhaps you could provide some traffic data to back up your speculation.
Regardless, I don't recall anybody saying no packets would be dropped. In fact, packets are discarded today in all packet-based networks, and will always be in the presence of congestion. Yet the world continues to turn...
And if they do drop proportionally more non-prioritized packets or tell non-prioritized senders to back off, then that is degradation of service and my question is, on what basis do you assert that that degraded service is "good enough?"
We're talking milliseconds here. That's well below the threshold of human perception, so your precious YouTube "best effort" videos will not look any crappier than they do now.
And non-prioritized senders are not told to back off. That's not how TCP or IP work. They merely note there was non-delivery of specific segments and retransmit. That is the way it's always been, and always will be.
Posted by: NTodd | July 10, 2007 at 12:29 AM
Please pardon my skepticism that any Federal regulatory body composed of people appointed by our current president is adequately protecting the public trust.
Then best of luck with any legislated regulatory regime.
"An ounce of prevention is worth a pound of cure." Ask Netscape's shareholders how they like their timely remedy in the MS antitrust enforcement.
Yeah, it would totally have made sense to break up Microsoft for tying a browser to Windows before there were such a thing as a browser.
Sorry, but having the legislature try to deal with this sort of thing is insane precisely illustrated by the MS/NS battle.
Posted by: NTodd | July 10, 2007 at 12:32 AM
Then best of luck with any legislated regulatory regime.
While it may be true that given regulatory capture the legislature is less responsive to the desires of voters than the desires of telcos, it's certainly harder for the telcos to manipulate the legislature and the justice deptartment than 5 mostly pro-big-business individuals appointed by one of the most pro-big-business presidents ever.
Posted by: smoothie | July 10, 2007 at 10:08 AM
That's just marketing speak and amounts to "trust me."
No, it's not. I've explained QoS technically, thanks.
No, you drew a picture which only astute readers who were following the comments between you and Alex (or people more knowledgeable about networking details) would realize was talking about separate data paths inside the router rather than, say, separate physical infrastructure on the egress interface.
What more sophisticated explanation do you require about QoS... Should I go into DiffServ marking and how that will impact CBWFQ at the ingress?
Yes, something like that would be nice, although my impression is that focussing on the ingress interface and ignoring the egress is potentially misleading; it's possible that I'm wrong, and if so I'd love to be corrected.
Yes. Thus, they can crank out packets at extremely high speeds on individual ports. Backbones are operating at OC-192 on single lambdas currently. Go ahead and try to choke that.
It's what happens at the transition between the backbone and destination networks I'm concerned about.
Are you asserting that all these routers do is rearrange packet order to smooth out bursts of non-prioritized traffic and never drop packets or send ICMP source quench messages? Seems unlikely.
Then perhaps you could provide some traffic data to back up your speculation.
You're the one who seems to have inside telco knowledge. I said it seems unlikely because absent dropping or source quenching of non-prioritized packets at points of congestion, queuing for the purposes of smoothing bursty non-prioritized traffic seems both unlikely to generate controversy and unlikely to generate much telco revenue. Congestion is where the QoS rubber hits the road, right?
Regardless, I don't recall anybody saying no packets would be dropped. In fact, packets are discarded today in all packet-based networks, and will always be in the presence of congestion. Yet the world continues to turn...
But they are not discarded right now on the basis of how much the transmitter payed the telco. The whole controversy is the perception that tiered service would change that. If you're saying that that is not the case, and have evidence that that is so, you've addressed my concerns. But your post did not say that.
And if they do drop proportionally more non-prioritized packets or tell non-prioritized senders to back off, then that is degradation of service and my question is, on what basis do you assert that that degraded service is "good enough?"
We're talking milliseconds here. That's well below the threshold of human perception, so your precious YouTube "best effort" videos will not look any crappier than they do now.
Retransmission due to packet loss causes the sender to back off. Retransmission due to repeated packet loss causes the sender to back off exponentially. I believe you know this, so I'm curious why you keep saying non-prioritized traffic won't degrade meaningfully.
And non-prioritized senders are not told to back off. That's not how TCP or IP work. They merely note there was non-delivery of specific segments and retransmit. That is the way it's always been, and always will be.
RFC 1812 says routers shouldn't send ICMP source quench. If tiering doesn't change this, great. That's the kind of information that I think was lacking in the rest of the post.
However, selectively dropping packets based on QoS parameters in the telco networks has a simliar effect (see above).
Posted by: smoothie | July 10, 2007 at 10:48 AM
it's certainly harder for the telcos to manipulate the legislature and the justice deptartment than 5 mostly pro-big-business individuals appointed by one of the most pro-big-business presidents ever.
You do realize there are Dems on the FCC, right? And that the FCC has already enforced "neutrality" when necessary, right? No? Hmm.
No, you drew a picture which only astute readers who were following the comments between you and Alex (or people more knowledgeable about networking details) would realize was talking about separate data paths inside the router rather than, say, separate physical infrastructure on the egress interface.
Oddly, blogging is about interaction. As I noted, I was oversimplifying so people would get an idea of what QoS is about, and Alex added some value to the picture. But that was not the only part of the discussion about QoS. And I also noted that this is a single infrastructure. So really, at this point, you're just trying to be an ass.
Yes, something like that would be nice, although my impression is that focussing on the ingress interface and ignoring the egress is potentially misleading; it's possible that I'm wrong, and if so I'd love to be corrected.
I wasn't implying I would focus on ingress only. Do you not get the idea of examples in a list? Egress is also important, but ingress is where decisions about which path to take are made, and where classification has the most impact across the provider network, which is what NN is all about.
It's what happens at the transition between the backbone and destination networks I'm concerned about.
It's easy: IP packets are handed to the customer's router, and it decides what to do based on customer policy. This also happens to be completely irrelevant to the NN discussion.
You're the one who seems to have inside telco knowledge.
Yeah, and? I'm supposed to dig up data to prove your point for you?
I said it seems unlikely because absent dropping or source quenching of non-prioritized packets at points of congestion, queuing for the purposes of smoothing bursty non-prioritized traffic seems both unlikely
Uh...queuing isn't unlikely. It actually happens now.
to generate controversy
Only because people are trying to invent controversy.
unlikely to generate much telco revenue.
As I already said, in the long-run it won't. And certainly we don't need legislation to make it so.
Congestion is where the QoS rubber hits the road, right?
Yes, and why we need QoS is because we're going to cram more types of data into the network. So we need to classify it so all apps get the QoS they require. This is becoming a recursive argument you're having.
Retransmission due to packet loss causes the sender to back off. Retransmission due to repeated packet loss causes the sender to back off exponentially. I believe you know this, so I'm curious why you keep saying non-prioritized traffic won't degrade meaningfully.
Yes, I do know this. I keep saying it won't degrade because it won't. Please reread the section about delay-sensitivity.
RFC 1812 says routers shouldn't send ICMP source quench. If tiering doesn't change this, great. That's the kind of information that I think was lacking in the rest of the post.
I didn't mention it because it wasn't relevant. You're now inventing objections to something that is not there, so I'm not going to anticipate every damned stupid argument.
Posted by: NTodd | July 10, 2007 at 11:47 AM
Adding: NOT classifying data will cause MORE problems. If you want to run different apps over any IP network, you're going to need QoS mechanisms. Customers are implementing this on their LANs now with VoIP, and providers need to be able to do the same.
To not classify at all is insanity. The biggest issue is whether a provider should be able to charge for it. They should because they're spending capital to deploy QoS-capable networks.
Posted by: NTodd | July 10, 2007 at 11:59 AM
Who's being an ass? You wrote:
Yes. Thus, they can crank out packets at extremely high speeds on individual ports. Backbones are operating at OC-192 on single lambdas currently. Go ahead and try to choke that.
and:
Yes, and why we need QoS is because we're going to cram more types of data into the network. So we need to classify it so all apps get the QoS they require.
So is or is not congestion an issue? You can't have it both ways! To whit:
It's what happens at the transition between the backbone and destination networks I'm concerned about.
It's easy: IP packets are handed to the customer's router, and it decides what to do based on customer policy. This also happens to be completely irrelevant to the NN discussion.
It's irrelevant? The backbone runs right to my door, does it? No telco router makes desicions to drop packets where the backbone transitions to slower networks that the telco, not the customer, control?
NOT classifying data will cause MORE problems. If you want to run different apps over any IP network, you're going to need QoS mechanisms. Customers are implementing this on their LANs now with VoIP
Yes, I know, I have implemented QoS at the ISP/corporate interface. I am not arguing against QoS. I'm all for teclo customers prioritizing packets based on their own data needs.
and providers need to be able to do the same... The biggest issue is whether a provider should be able to charge for it. They should because they're spending capital to deploy QoS-capable networks.
Or, they could just continue over provisioning and let the individual customer do QoS, dropping packets that matter less to them, not which come from senders who paid the telco less.
The reason people don't want the telcos to treat the backbone soley as their property is that it's also vital infrastructure. I'd even be open to public subsidies to help defray the telcos' infrastructure costs if they weren't already some of the most obscenely profitable corporations on the planet!
Oddly, blogging is about interaction. As I noted, I was oversimplifying so people would get an idea of what QoS is about... So really, at this point, you're just trying to be an ass.
And I thought (and still think) your oversimplification was misleading and that my comments raised some legitimate points. Your hostile tone suggests you don't like non-fanboy interaction, so I'll quit now.
Posted by: smoothie | July 10, 2007 at 12:31 PM
Hmmm.
On re-reading your whole post I have come to the inescapable conclusion that I was being a bit on an ass! Particularly since you did in fact admit it was a gross oversimplification and that and that there was in fact one infrastructure.
I stand fast on my assertion that the telcos shouldn't be in the business of dropping packets based on cost but rather should focus on overprovisioning (and stick to charging us all based on total bandwidth).
Posted by: smoothie | July 10, 2007 at 03:55 PM
Who's being an ass?
You, for trying to create a controversy that doesn't exist in a technical environment at all.
Yes. Thus, they can crank out packets at extremely high speeds on individual ports. Backbones are operating at OC-192 on single lambdas currently. Go ahead and try to choke that.
and:
Yes, and why we need QoS is because we're going to cram more types of data into the network. So we need to classify it so all apps get the QoS they require.
So is or is not congestion an issue? You can't have it both ways!
I'm not having it both ways. There's enough speed for all apps to be routed through the network with the appropriate QoS parameters required. That's the point. The router can crank out packets at wireline speeds, and it prioritizes out those serial ports so that all packets get fair access to bandwidth according to their specific requirements. It's what ATM has been doing for ages, and now IP can with the help of DiffServ and MPLS.
It's irrelevant? The backbone runs right to my door, does it?
Yes, it's irrelevent to NN because the whole issue is surrounds prioritizing within the backbone, not on the customer prem. And the backbone does not run to your prem--that's the access component access, not transport (not unlike the whole issue of copper lines from your house to the CO).
No telco router makes desicions to drop packets where the backbone transitions to slower networks that the telco, not the customer, control?
I don't even understand this question to be able to answer it. If the customer's own netwok can't handle the traffic, it's not a provider issue (unless that customer would like to buy managed network services past the traditional demarc, which they can). The backbones purpose in life is deliver stuff from edge to edge.
Or, they could just continue over provisioning and let the individual customer do QoS, dropping packets that matter less to them, not which come from senders who paid the telco less.
Without QoS, providers would have to build out as much as 5 times more, which costs more money that you won't let that recapture.
The reason people don't want the telcos to treat the backbone soley as their property is that it's also vital infrastructure. I'd even be open to public subsidies to help defray the telcos' infrastructure costs if they weren't already some of the most obscenely profitable corporations on the planet!
Telco revenues are in the shitter, particularly wireline. The only places they can make it up right now are broadband and wirelss. To deny them revenue sources to maintain the networks they invest in is to shoot that network down.
The phone infrastructure is also vitally important, yet nobody questions phone companies' right to create new services.
I stand fast on my assertion that the telcos shouldn't be in the business of dropping packets based on cost but rather should focus on overprovisioning (and stick to charging us all based on total bandwidth).
They're already in the business of dropping packets. That's the nature of any shared infrastructure. What they're doing is providing more guarantees for things that are more sensitive to loss and delay. The very apps people are demanding be run over the Internet. To hold fast to a position that the Internet shouldn't evolve the way private IP networks have is to ignore reality as much as Bush ignores that the surge isn't working.
But yeah, I'm glad you see my point about being an ass. Of course, I'm incivil in general, so you should always take that with a large grain of salt. Ask Bill.
Posted by: NTodd | July 10, 2007 at 05:41 PM
Without QoS, providers would have to build out as much as 5 times more, which costs more money that you won't let that recapture.
Who said I wouldn't let them recapture it? There are other billing models they can use, including some that they use right now, like charging for aggregate bandwidth usage instead of just capacity. Consumers don't tend to go for that, but plenty of businesses do.
To hold fast to a position that the Internet shouldn't evolve the way private IP networks have is to ignore reality
Don't conflate "the internet evolving" with a particular telco pricing scheme.
Posted by: smoothie | July 10, 2007 at 07:38 PM
Who said I wouldn't let them recapture it? There are other billing models they can use, including some that they use right now, like charging for aggregate bandwidth usage instead of just capacity. Consumers don't tend to go for that, but plenty of businesses do.
You appeared to be arguing against tiered pricing. And yes, there are plenty of models. Which they should be allowed to determine without regulatory interference. One way is to establish a higher pricing tier for premium services such as one that would allow customers to classify their traffic.
Don't conflate "the internet evolving" with a particular telco pricing scheme.
I'm not. I'm saying QoS is a necessary evolution, and the telcos have a damned right to charge for it somehow.`
Posted by: NTodd | July 10, 2007 at 11:45 PM