Free Press wants the FCC to mandate a dumb Internet
- Part 1 – Free Press Net Neutrality proposals would devastate economy
- Part 2 – Free Press wants the FCC to mandate a dumb Internet
- Part 3 – We can’t pretend wireless and wired networks are the same
This is part 2 of this series analyzing Free Press’ 198-page reply to the FCC that recommends more restrictive Internet regulations. In part 1, we talked about how Free Press’ proposals would devastate the U.S. economy by precluding ISPs from premium services and neutering their existing television and phone businesses. In this segment, we will analyze Free Press’ recommendation that the FCC mandates a dumb Internet that is hostile to significant classes of Internet applications.
To prove their point, Free Press primarily relied on a previous Free Press paper from Chris Riley and Robb Topolski paper about “The Hidden Harms of Application Bias” which we thoroughly debunked here and here. Without rehashing every detail of the entire debate, we’ll summarize the issues here.
The Riley-Topolski paper made the following key assertions
- Prioritized networks degrade low priority applications
- Prioritized networks degrade overall network performance
- Locks the Internet into typical usage patterns of 2009
- Application bias has not shown to be necessary or even substantially beneficial
Why the Riley-Topolski paper is wrong
- Prioritized networks (outside of managed services) only give priority to lower bandwidth applications that are sensitive to latency and jitter. No matter how much the network prioritizes Voice over IP (VoIP) or online gaming, VoIP and gaming will never use more than a small fraction of the bandwidth used by peer-to-peer (P2P) file transfer. VoIP and gaming generally occupy less than 0.1 Mbps out of the average broadband connection of 4 Mbps downstream and 1 Mbps upstream in the United States, which means they get no more than 2.5% of the downstream network and 10% of the upstream network.
- This assertion is the opposite of reality and it was proven wrong by test data. Furthermore, why would network operators implement a product that makes their network less efficient? Cisco has stated before the FCC QoS workshop that they can’t sell a networking product that doesn’t have prioritization technology because their customers value the enhanced efficiency of products with prioritization capability.
- QoS technology evolves with everything else and it will remain accurate like it is today or else no one will use it. The risk of misclassification or application cheating is a non-issue because there is no benefit to cheating. Even if a high bandwidth application manages to fool a router into giving it “priority”, high priority merely grants low latency and not high bandwidth. Network devices like routers and switches can configure the high-priority low-latency lane with a low bandwidth ceiling so that priority cheating is counterproductive to high bandwidth applications.
- This goes against our best knowledge of network engineering. Prioritized networks that protect latency sensitive applications allow high bandwidth applications to reach their full potential because they need not fear the repercussions. Without network protection for latency sensitive applications like VoIP and online gaming, users who also uses peer-to-peer (P2P) will typically shut down P2P while they use VoIP or online gaming. The more P2P users shut down, the fewer the number of computers that contribute to the seeder pool that is crucial to all P2P networks.
The fallacy of the dumb network religion
The fundamental problem with Free Press’ position is that they have put all of their misguided faith into the dumb network which they refer to as a “First In First Out” (FIFO) network. Free Press claims that the Internet has always been a FIFO network (which is completely wrong because the Internet accommodated and even standardized out-of-order operation from the beginning) and should remain a FIFO network. The idea is that a FIFO network operates on a first come first serve basis and rules out any potential mischief on the part of the network operator. The belief is that the FIFO network operates in some kind of enlightened state without “application bias”.
The fallacy of this view is that Internet Protocol (IP) networks are fundamentally biased, unfair, and unbalanced towards certain classes of applications such as VoIP, online gaming, and video conferencing. The main purpose of network prioritization is to undo this fundamental bias so that all applications get fair treatment from the network and that all applications work well.
To illustrate how ridiculous a FIFO network is, we can look at what happens when a P2P application and VoIP application shares a broadband connection. The FIFO network operates on a first come first serve basis which sounds fair but really isn’t. That’s because P2P applications can literally stuff hundreds of packets into the network operator’s switch before a single VoIP packet makes it to the switch. So by insisting FIFO operation, the network has to forward hundreds of P2P packets (simply because they showed up first) before it forwards a single VoIP packet. The wait time forced on the VoIP packet is called jitter (explained here with animations) and it causes the VoIP packet to expire resulting in audio dropouts in the phone call. This isn’t even remotely fair, neutral, or efficient and it is simply a dumb way to run a network.
The fair way to operate the network is for the switch to sequentially forward hundreds of P2P packets or as many as necessary so long as nothing else wants to use the network. If a few individual VoIP or gaming packets show up and the waiting line is filled with hundreds of P2P packets, the network will bump those VoIP and gaming packet to the front of the line and eliminate the wait times. The result is that P2P will continue getting the lion’s share of network bandwidth despite its low priority status on the network switch but it will not be allowed to induce packet expiration on VoIP, online gaming, or other latency sensitive applications.
[...] Part 2 – Free Press wants the FCC to mandate a dumb Internet [...]
(answering 1-4 above)
1. Then “Topolski-Riley’s” (our) assertion remains true, or at least is not debunked. You say the effect is small, but you do not demonstrate that there is no effect.
2. If we said that over-correction of latency can cause effects enough to be problematic for buffered video streaming, we didn’t test that (and I don’t believe it, nor do I believe that we said precisely that — you’d pretty much need to stall the buffered video download to a rate slower than playback). Beyond that, the whole argument made in today’s article and the past one relies on the reader conflating network prioritization (between triple-play services) with ISP-directed prioritization between different types of traffic on the general-purpose Internet. Very different things.
3. There are many useful applications for QoS, including user-directed prioritization that is permitted (or encouraged) by the Internet Standard RFCs. The balance belongs in private networks or on carefully subdividing multi-purpose networks between managed services and the all-purpose Internet.
4. ISP’s application bias locks in today’s Internet at the expense of tomorrow’s choices. ISPs should not fear that user-directed prioritization that I mention above. Generally, users will choose wisely (it is in their best interest to do so, as you allude to). The choices they would (en masse) make harmonize well with the choices ISPs might force if they were allowed to do so.
Response to Mr. Topolski below.
1. The mere reduction of bandwidth for P2P does not constitute “harm” unless you are of the opinion that the network should only allocate bandwidth to P2P and nothing else. It would only be harmful and unfair to P2P if the network allocated more average bandwidth to other applications than P2P, but this is not the case and P2P will continue getting the highest average bandwidth even when P2P is given the lowest priority. Other application such as web surfing or HTTP buffered video streaming deserves an equal share of bandwidth as P2P yet P2P hogs more than 90% of the bandwidth on a dumb FIFO network. If the network prioritizes web surfing such that it now gets 50% of the bandwidth on average, that would be perfectly fair. This is not the kind of blunt and harmful hard-limit throttling on P2P that some of the Canadian ISPs are using (which the CRTC, their version of the FCC allows).
Furthermore, a network which prioritizes VoIP over P2P does not change the allocation of bandwidth. VoIP still gets 2% while P2P gets 98% with or without prioritization? The only difference is that the prioritized network eliminates jitter for VoIP.
2. “If we said that over-correction of latency can cause effects enough to be problematic for buffered video streaming,” You did say that, and it was wrong. There is no such thing as an “over correction of latency” that would cause excessive jitter beyond a few hundred milliseconds much less 5000 milliseconds needed to starve out buffered video stream. To be more precise, correcting latency for a 218-byte VoIP packet by pushing it to the front of the line ahead of the video streaming packets would cause an extra 0.58 milliseconds of delay for the video stream. That’s thousands of times under the threshold for starving a buffered video stream.
3. User-directed prioritization is good, and I have stated on many occasions that it should be given preference over the ISP’s default settings so long as the user isn’t abusing their priority budget, but that does not mean we shouldn’t have ISP default settings. This is because most users and most applications don’t bother “directing” the network and this is the current status quo on most networks. The end result is that real-time applications like VoIP and online gaming suffer badly at the hands of P2P and to a lesser extent aggressive YouTube cache ahead buffering behavior.
4. Again, you cannot be presumptive about how tomorrow’s QoS configuration or technology. Technology always changes with the times to meet the needs of the user. More importantly, the type of “bias” I am advocating is a counter-bias that stresses fairness across all applications such that one application doesn’t get to monopolize the network for hundreds of packets in a row while starving other applications. The type of prioritization I am advocating says that all applications should at least get an equal chance at getting their packets forwarded. This philosophy of engineering networks such that they’re fair and hospitable to all applications doesn’t ever expire and it has always been the philosophy network engineers have strived for.
(more 1-2-3-4)
1. Prioritization does nothing unless there is queuing, at which time it reorders the queue placing the preferred traffic first and, if there is any room left, stores the non-preferred traffic. The non-preferred traffic that is queued is delayed (a harm). Any remaining non-preferred traffic is dropped (a harm). Because less preferred traffic is being dropped, the allocation of one type against another -IS- changed.
We both agree that in the battle of VOIP versus P2P, that’s good. We may both design our private networks this way. We may both decide to admit upstream traffic to the Internet in this way. However, it’s not the ISP’s place to make those decisions for us. The ISP doesn’t have a way to positively know “real-time” P2P versus passive downloading. Its method of guessing (DPI) is both flawed for today and harmful to innovation tomorrow.
2. I deny saying it. That said, an overcorrection is absolutely possible when the sending system is sending out 30-50 VOIP packets a second in order to interspace VOIP packets between larger packets (much like some of your drawings have shown, except that your drawings are a proposal of something that already NORMALLY happens with some VOIP software on a non-prioritized network). Prioritizing that over other traffic that was providing the interspacing would cause the VOIP packets to be crowded together and would INCREASE jitter.
3. We probably agree more than disagree here, except for the conclusion. Again, I maintain it’s not the network’s job to fix it and it’s harmful to give the job to the network (especially in non-standard ways) when the end points already own it.
4. Take it to the IETF, where it belongs. When it passes, I’ll have no choice but to salute it. Meanwhile, what hundreds of millions of Americans want is Internet access, not “Comcast” or “AT&T.” They overwhelmingly want the Internet more than they desire the features or weirdnesses added by ISPs.
1. Prioritization in the case of favoring low-bandwidth applications will usually never change allocation of bandwidth unless the high bandwidth application was 100% aggressive (usually not the case because it would cause too many problems). So in practical conditions, something like P2P will continue getting 98% (typically less since it’s hard to find enough generous seeders) and VoIP will get its 2% which means allocation won’t be changed. But again, even if allocation ratios are changed to 50/50, that would be perfectly justifiable. What does change under prioritization is that VoIP no long faces any jitter.
The ISP absolutely has a duty to fix these unfairness problems since users cannot fix the downstream jitter problems by themselves. They might be able to limit the P2P downstream performance and reduce the jitter problems, but that needlessly sacrifices P2P performance while only having marginally beneficial effects on downstream jitter. When the ISP gives a priority queue to VoIP, gaming, and low-duration HTTP, P2P can run at full throttle and the effects of jitter can be almost completely eliminated.
You are simply wrong that we can’t positively identify and prioritize low bandwidth applications and you’re wrong that it might induce app developers to masquerade their packets. Applications can’t hide how bursty they are and they can’t hide how much bandwidth they use.
The fact is that you don’t even understand that we’re not talking about DPI in relation to network management. DPI isn’t this monster you portray it to be. I’ve made this very clear at http://www.digitalsociety.org/2009/10/understanding-deep-packet-inspection-technology.
2. You can deny all you like, but I quote you here (http://www.digitalsociety.org/2009/11/debunking-the-myth-that-prioritized-networks-are-harmful) saying that prioritization can cause problems for buffered streaming video.
Your claim that prioritizing VoIP will cause problems for other applications is incredibly ignorant. Priority “overcompensation” (the proper term is “queue starvation”) on lower priority applications is one of the first things network engineers are taught to *avoid*. It was possible when mis-configuring older priority queuing algorightms when priority is given to bursty or high bandwidth applications without adequate restraints, but even a mis-configured system would not have problems when prioritizing VoIP traffic.
This is because VoIP traffic fundamentally is self constrained to 50 packets per second at 218 bytes per packet whereas a P2P application might want the maximum 500 packets per second at 1472 byte per packet on a 6 Mbps link. And because those VoIP packets are evenly spaced rather than all packed together back-to-back, there is no possibility of inducing jitter on other applications.
The reality is that modern queuing algorithms default to a configuration that avoids queuing starvation on all the queues so there’s a much lower chance of operator error.
3. I’ve already explained that the individual and end points are not technically capable of implementing a solution for downstream jitter. The ISP’s switching infrastructure before the last hop is the only place where this downstream jitter can be completely eliminated and not just reduced, and jitter is one of those things were we want to get rid of it and not just lower it.
As for upstream jitter, a very small percentage of the population would even have the equipment or knowhow to address this issue so it’s very reasonable for the ISP or the VoIP provider to pre-configure this for the end user. In fact, this is precisely what companies like Vonage or Lingo do today when they preconfigure their VoIP adapters to favor their own VoIP traffic above all other traffic.
4. I absolutely agree that the IETF is where this discussion belongs and has been discussed since the beginning of the Internet. It does not belong in a Free Press advocacy piece asking the FCC to ban prioritization technologies inside the network and it certainly does not belong in FCC regulations or congressional legislation. The IETF should not have to ask the FCC or US Congress permission to implement a standard update.
[...] Part 2 – Free Press wants the FCC to mandate a dumb Internet [...]
[...] [...]
[...] [...]
Leave your response!
Twitter Feed
About Us
Digital Society is a digital think tank that believes culture and commerce are inseparable, that the digital economy flourishes when people are free and rights are secure, and that free markets free people.
Digital Society is an independent 501(c)3 non-profit organization, funded by donations from Jon Henke and from Arts+Labs. We advocate for a pro-culture, pro-commerce digital society through research, analysis and debate on emerging technology issues.
Reply Comments
Transparency and interactivity are trademarks of the Internet era, and we aim to foster them here at Digital Society. It is inevitable that some people will disagree with the technology policy positions we take. We want to have that constructive debate.
The Reply Comments feature gives our critics a chance to respond to our viewpoints and the Digital Society audience convenient access to competing arguments. Any time we directly challenge the views of an individual or a group on this site, the party in question may substantively respond in a guest post.
Please contact executive director Jon Henke by e-mail.
Subscribe
Recent Posts
Recent Posts
Most Commented
Most Viewed