In a message written on Mon, Nov 29, 2010 at 10:22:34PM -0500, Christopher Morrow wrote:
see craig's report from nanog47: <http://www.nanog.org/meetings/nanog47/presentations/Monday/Labovitz_ObserveReport_N47_Mon.pdf>
not for a time has Comcast been solely an 'eye-ball' network... or so they think.
I think you are misreading the data. From googling around it appears there are somewhere between 90 and 100 million "broadband subscribers" in the united states. Comcast claims to have somewhere between 15 and 17 million broadband subscribers, and they are the largest cable company in the US. With around 18-20% of all broadband end-users in the US Comcast, if you believe Arbor's numbers, generate 3.12% of all Internet traffic. Comcast also sells business service (not cable modem, but like GigE to the prem) which is propping that up a bit. If the FCC wanted to do something useful they would look at the combined ratio of all /customers/ of an ISP, and then require their peering policy to allow for around 2x of that. For instance, if you summed all Comcast customers and did the ratio of out:in and got 3:1, they should at a minimum be required to peer with someone at 6:1 IMHO. I have no idea in Comcast's case specifically, or in any recent case as my skin isn't in the game right now. However I am quite sure in the past I have delt with networks who wanted 2:1 on peering, but where I was nearly positive their customer base was 3:1 or 4:1. Basically the ratio became an excuse to depeer anyone they didn't like, it was all a sham. While I think ratio requirements are just plain stupid, I do think it needs to be considered when looking at peering. If you do hot potato routing the person on the "wrong end" of the ratio ends up carrying the traffic longer distances. If you look at long haul bandwidth on a bit-mile basis this can be unfair in some circumstances. The thing is though it's easy to fix. Networks could use MEDs (yes, they work on Internet scale routing), selective leaking (w/no-export), peering with regional ASN's (many of the large eyeball networks are subdivided internally) or any number of other very simple configurations to balance this issue. But I come back to my fundamental beef with cable and DSL providers, when you're selling 50/5 (10:1 ratio), 25/5 (5:1 ratio), 12/2 (6:1 ratio) services, you can't expect to maintain a 2:1 or 3:1 ratio with your peers. If you look at the TV side of the business the eyeball network is the whipping boy 99 times out of 100. Look at the recent Fox v Cable Vision dispute, Cable Vision caved. Users want content, users pay Cable Vision, Cable VIsion gets millions of angry calls, Fox runs a few ads how Cable Vision is the big bad guy and they have deals with everyone else. Go back to previous cases, almost always the eyeballs cave. Provides are trying to change this in IP space, because they don't like it. They want Netflix/Amazon/Apple/RIAA/MPAA to pay, and not be in charge. For the moment this works, if Netflix can't deliver via the Internet their users just request DVD's in the mail; a peering spat hurts Netflix more than Comcast. But, as users cut the cord, and get more of their content over the Internet I think we'll see the same shift. Outside of Nanog Ma and Pa Citizen don't even know what the word peering means. All they know is when they can't get their Netfix streaming to work they call their provider and complain, possibly going as far as to switch services. Now, while it may seem I'm taking Level 3's side of this dispute I am not. Sadly when these things spill out in public like this it is generally because both sides have been acting like idiots with each other in private for months or years. Maybe Level 3's been a model citizen in this case and has been wronged, but I doubt it. The problem is it all happened in private, and nice press releases from both parties aside we really have no idea what happened behind closed doors, who asked for what, who's egos got out of control, etc. I'm not going to call a winner or a loser, just point out how broken some of the arguments put forth are..... -- Leo Bicknell - bicknell@ufp.org - CCIE 3440 PGP keys at http://www.ufp.org/~bicknell/