Elaine C. Kamarck writing an op-ed piece in The Boston Globe has is exploring how to fund the required build-out of Internet capacity. (Hat tip: Andy Abramson) I have never fully understood the peering business in Internet and this column has not helped much. So this gives me an opportunity to raise some questions.
This is my understanding (mostly based on Tom Evslin’s multiple posts on this topic) of how peering works in the Internet: Peering between two ISPs take place when one delivers traffic to another when the former requires the help of the latter to carry the traffic further in the Internet towards the intended destination. If historically the two ISPs exchange comparable amount of bandwidth, then each carry others’ traffic with no further monetary exchange; otherwise there is a pairwise business agreement on how much to charge for the excess traffic. So far so good. The first confusion comes when people say that Internet peering uses a different model than PSTN. In my understanding, this IS the PSTN model. So what am I missing?
Since most of the current crop of Internet applications consume bandwidth asymmetrically. By that I mean, the consumers generate very little bandwidth towards the Internet, but consume inordinate amount of bandwidth from servers at the other-end of the connection. This means that when the access providers like telcos and cablecos must be receiving revenue from the ilk of YouTube/Google. If so why do they place a limit on the amount of downlink bandwidth consumed by their users? Also, shouldn’t this be enough of an incentive for access providers to make sure that the backbone has sufficient capacity to support all these applications?
If the peering deals only with the consumed bandwidth, why should we worry about which application generated bandwidth? Even the blindsided PSTN didn’t/couldn’t charge a different rate for fax/modem call. In my opinion the solution lies in letting the consumers generate bandwidth and the access providers monetize the uplink traffic. This means as a first step, the access link should be symmetric; after all the backbone is.
Posted by aswath at December 26, 2007 03:58 PM
Peering is run on an essentially ad hoc basis, with nothing as sophisticated as the interconnect billing systems used by telcos. Either the two ISPs consider themselves as being on the same tier and agree to peer, or one ISP considers it is on a higher tier than the other and will ask for transit fees (fluctutate according to market rates, currently about $100/Mbps/month).
All self-respecting Tier-1 transit ISPs (a very rarefied group which invludes AT&T and Verizon but none of the cablecos) run non-blocking backbones where there is sufficient overengineered capacity on all links to handle traffic. The only situation where a link will saturate is during transient failures which cause routing sub-optimality.
Restrictions on upload speeds serve two purposes. For cable Internet providers, the CATV medium has limited spectrum and cablecos with copper coax plants have reached capacity, which already limits their ability to offer HDTV channels until the analog spectrum is released circa 2010. Thus they would rather not provision channels for upstream traffic (specially P2P) if they can help it.
For DSL suppliers, setting assynnetric speeds allows them to offer higher download speeds, which helps with marketing if not in terms of actual performance. Not all DSL is assymetric - SDSL is symmetrical and essentially a replacement for T-1 service. Limits on upstream/downstream on DSL providers are mainly a way to permit price discrimination between home and business users, as are static IP addresses.
Per-application pricing has no connection whatsoever with the costs, it is just a persistent pipe dream of telcos to go back to the bad old days where they could charge whatever the market could bear, secure in their monopoly. It's as if your phone company charge you for an emergency cal to the hospital at 10 times the going rate because that call is more valuable to you.
Posted by: Fazal Majid at December 26, 2007 07:16 PMFazal:
Thanks for the clarifying comments. Ad-hoc or not, I feel they are essentially the same in the sense that there is a fee for excess traffic. Since currently the traffic is asymmetric in Internet, the "little guy", consumer access providers are really the big guys because the excess traffic in in their favor.
I still can not think that the telcos dream of charging by the application. They didn't do it in the voice world - they didn't tell an investment firm that they will charge more to carry their calls because they generate more revenue from the call.
Posted by: Aswath at December 26, 2007 07:40 PM