Say you need to reach www.rabelaisian-wit-is-pretty-dirty-stuff.com, and the relevant Web server sits in a Vladivostok data center. But Hyperlocal Internet, your Internet provider, has no direct connection to the Vladivostok hosting company's Internet provider. So how to send your request for a Web page across the Bering Sea?
Internet transit provides the answer: one ISP pays another well-connected network to deliver Internet traffic to networks with which the first ISP has no direct peering connection. (Read our primer on peering and transit.) Typically, such transit deals have been priced at a "blended rate" under which the transit provider charges a flat price per Mbps of connectivity; in other words, the transit providers charges for the size of the pipe it provides, regardless of how far the traffic is going or how high transit demand is at the moment.
To reach the Vladivostok ISP, Hyperlocal's transit provider might need to haul those bits across the US and perhaps over the Pacific before handing them off to another network in, say, Singapore, that can get them further along the way (this is called "off net traffic"). Such traffic imposes higher costs than if Hyperlocal's data is destined for one of the transit provider's own customers in Omaha (called "on net traffic"), for instance—yet the costs to Hyperlocal are the same either way with a blended rate.
A new paper (PDF) called "How Many Tiers? Pricing in the Internet Transit Market" from Georgia Tech and Stanford researchers Vytautas Valancius, Cristian Lumezanu, Nick Feamster, Ramesh Johari, and Vijay V. Vazirani argues that this approach is inefficient.
"If network costs are highly variable, less costly flows in the blended-rate bundle subsidize other, more expensive flows," they note, and then point out that transit providers have already started moving away from this model, offering tiered pricing instead.
Such pricing might change based on traffic's cost—whether particular traffic is on net or off net, for instance. But transit providers are leaving cash on the table with this simple approach to tiering: "Our results show that the common ISP practice of structuring tiered contracts according to the cost of carrying the traffic flows (e.g., offering a discount for traffic that is local) can be suboptimal."
To capture the full value of transit traffic, ISPs could attempt to bill every single traffic flow—but this poses huge challenges in terms of complexity and the resulting overhead. Also, it might be baffling to customers, who would be at a loss to estimate how much money they might owe next month in transit under such a scheme.
But mathematical analysis in the paper shows that transit providers don't have to worry that they're leaving mad cash on the table by not resorting to this sort of complex billing, however. "Our analysis shows that, indeed, in many cases, an ISP reaps most of the profit possible with infinitesimally fine-grained tiers using only two or three tiers, assuming that those two or three tiers are structured properly," says the paper. ("Properly" in this case meaning pricing that considers both demand and cost-to-deliver at the same time.)
ISPs have started to implement such strategies, though details are hard to come by—transit and peering agreements have always been guarded closely. But the shift away from simple blended rate pricing shows the Internet interconnection market maturing, according to the paper's authors, and it results in a more efficient market.
That market certainly has been a brutal one for providers, with costs for transit falling by an average of 30 percent a year. Sadly, when it comes to home Internet pricing, transit price drops don't often translate into lower monthly bills.
Listing image by Photograph by Julian Burgess
reader comments
54