February 28, 2008

Pooper scooper

I’m just having some fun taking a Waitrose Colombian coffee flavoured milk drink break, reading David Reed’s submission to the recent FCC hearing. It’s all about Comcast’s approach to squishing BitTorrent traffic on its broadband network. (Sorry, no link, I’ve only got a copy via email.)

I’m not in favour of network neutrality regulations, and feel that Comcast did a Bad Thing™ that should be punishable under existing consumer protection, contract and tort law.

In his submission, David puts forward several ideas. The first says:

All it takes to be part of the Internet as an Autonomous System is to agree to participate according to the very simple ground rules of the Internet.

Bzzt. There is no social or legal contract to adhere to the IETF’s recommendations. To me this feels like communitarian nonsense (in the nicest possible way, separating the argument from the arguer). I think this confuses what should be (maximising total utility through co-operation) with what is (a prisoner’s dilemma-style game where cheating can be rewarded). This is entirely a private matter between Comcast and those other networks it interconnects with.

Somehow, the dignified manner in which IETF standards are arrived at is supposed to re-inforce this point of view.

These congestion control techniques can only work well if they are standardized across the entire Internet. New techniques are introduced carefully, typically orchestrated in the Internet Engineering Task Force, which is a collection of engineers and researchers who resolve these issues Internet-wide, independent of the vendors and operators, but taking their needs seriously.

No, I’m more than happy for ISPs to experiment with techniques that the IETF thoroughly disapproves of. Central planning, even that of benevolent technocrats, is fraught with danger.

A more convincing argument comes later:

Comcast used these non-standard mechanisms in an unexpected way, potentially disrupting systems and applications that are designed assuming the expected behavior of the Internet.

It’s not what you’d write in a public submission, but Comcast’s actions is a bit like a dog poo in the public park. Not only does it offend those who tread in it, but now everyone else has to expend time and energy looking for poo and telling their kids to keep away from the grass near the entrance. There’s an externality which is imposing costs on application and device designers everywhere. I’m happy for us to have regulations that manage pollution, but that carefully balance the true costs of the pollution against the immediate benefits of production. If in doubt, regulate less. Indeed, in this case, probably not at all, since if a legit P2P application was being messed up, Comcast customers would either churn or be willing to pay less.

And the real reason to be furious at Comcast comes at the end:

When Comcast or any Internet Access Provider claims to offer Internet Access, they implicitly agree to participate according to the standard practices of the Internet as a whole. Otherwise, all they may claim to offer their customers is “selective access to part of the Internet’s capabilities”.

Now, we might disagree what those “standard practices” are, but to me it’s limited to:

  • social measures to manage traffic (pick up the phone and tell the customer they’re over their fair use limits), or
  • applying traffic shaping to users with clearly defined and transparent policies and a system of individual notification of any special throttling,
  • dropping packets until the pain to that user modifies their behavior.

Event if RST injection (i.e. automated lying) is superficially the most efficient way of damping the traffic, users will respond to less technically efficient (but more socially acceptable) incentives. Yes, DOCSIS cable networks have a whacky uplink architecture that makes them very sensitive to lots of TCP/IP sessions being started at once, such as BitTorrent attempts. Just start chucking away 99% of their packets, irrespective of protocol, to reflect the social cost to the other users of uplink congestion. They’ll soon get the message and change when their Web browsing slows to a crawl. At the very least send them an email saying you’re throttling them, and why.

Picking on specific protocols, without telling the user, and then denying it, just isn’t part of the game. Particularly protocols that compete with your own video distribution network. Time for the referee to call a foul, and Comcast to scrape the sh*t off everyone’s shoes.

UPDATE: Before anyone gets too excited that Martin’s gone all neutralist, the offence is one that’s of the poop in the park magnitude, not that of premeditated murder that some more hysterical accounts would have you believe. Antisocial, irresponsible, stupid — but not criminal.

Posted by Martin Geddes at 06:30 PM
Trackback Pings

TrackBack URL for this entry:
http://www.telepocalypse.net/cgi-sys/cgiwrap/mgeddes/MT/mt-tb.cgi/908.

Comments

I like the dog poop metaphor, despite its graphic nature, because it seems to me you've hit on the problem: Comcast should be free to choose to do whatever it wants with its own network. However, if they choose a behavior that disrupts higher-level network, they could face sanctions from other networks that peer with them; they could be discriminated against, and, to follow their argument, there should be no recourse for Comcast being discriminated against by peers (or even individual Web sites).

I completely agree that there are no agreements binding the Internet. AS's to agree to certain things with IANA (very limited things, mostly about numbering) to obtain their status. But everything else is a matter of contracts among peers, which can state anything legal and enforceable -- or even unenforceable and illegal, and let the courts sort that out.

The Internet is an anarcho-syndicate, I've always argued. Independent entities agree specifically to certain rules that both avoid pooping in the pool, and allow them to ban those who do poop.

Posted by: at February 28, 2008 06:58 PM

Churn where? To dial-up? That's my alternative to Comcast in my neighborhood (almost, we just got ADSL with Ma Bell, but at 18,000ft from the CO it's nothing to brag about). Think DOCSIS uplink is bad, try the multiple TCP starts on DOrA. Time them right and get some amazing idle mode thrashing. Anyways, I think this is a lot of people venting their frustration over the duopoly status quot. We need some serious LLU competition, and the FCC has a loosing track record on delivering.

Posted by: at February 28, 2008 10:54 PM

I remember reading that increasing capacity would cost less than packet filtering. Is it true? I suspect it is.

Posted by: at February 29, 2008 04:42 PM

All it takes to be part of the Internet as an Autonomous System is to agree to participate according to the very simple ground rules of the Internet.

Bzzt. There is no social or legal contract to adhere to the IETF’s recommendations.

I think you've missed the point. There's no social or legal contract in the same way that there's no social or legal contract in a game of football. But if you want to play football you have to abide by the rules or else other people won't play with you.

This was especially true back when the internet was limited to large institutions and the telcos were only there to provide the pipes. If one institution decided to not play by the rules then other institutions would stop talking to them. The telco wasn't really involved.

I think this is the fundamental cause of all the debate over network neutrality. There's the traditional view of the internet as a network of equal peers and the new view where most people on the internet are individual customers of a service provider (and are therefore not equal peers, though maybe some of them are not be being made aware of this by their service providers).

The thing is, I don't think "participate according to the very simple ground rules of the Internet" really applies in the Comcast case. As long as Comcast are playing by the rules in their external connections then they are being neutral according to the traditional view. This is the same as how way back at Uni, as long the University played by the rules with the internet at large the internet at large could care less what happened inside the university network. The Uni was the equal peer, not me. With Comcast, they are the equal peer, not the consumers of their service.

Posted by: at February 29, 2008 06:30 PM

Reed's an interesting character, but his FCC filing and hearing comments were more street theater than serious technical analysis. Comcast's problem is a small number of users who have effectively monopolized their limited upstream bandwidth, at the expense of the web surfing satisfaction of the other 95%. They deal with this imbalance in bandwidth appetites by adjusting the amount that P2P has as a whole under conditions of higher than normal load.

Comcast screwed up in not disclosing fully what they're doing and why, but as a technical matter their approach is sound enough.

IETF people have a long-standing grudge against middleboxes injected RST packets, as you can see from RFC 3168 (ECN for IP), but it's silly. RSTs are good because packet drop is inadequate, which is why ECN was invented in the first place.

Reed is simply reflecting that and adding a little "Respect My Authorita" as a second-rate Eric Cartman.

Posted by: at March 17, 2008 07:18 PM
Please enter your comment below. Your comment will not appear immediately -- they all go for pre-approval by me because of the volume of spam I receive.







Remember personal info?