March 1, 2014 at 12:25 PM by Dr. Drang
Last weekend’s deal between Netflix and Comcast has been discussed to death, so you might consider this an exhumation. Most of the early reaction was not at a particularly high level: “Deal bad. Comcast bad. Net neutrality good. Hulk smash!” My knee jerked this way, too, but it was put back in its joint by more nuanced views, in particular Kevin Drum’s and Ben Thompson’s. Their articles got me thinking about railroads.
The key issue in the Netflix/Comcast matter is the amount of traffic Netflix dumped on Comcast through its carriers, Cogent in particular. Here’s the nut of Kevin Drum’s explanation:
In the past, big internet backbone carriers all interconnected to each other free of charge. Maybe CNN connects to the internet via AT&T and the New York Times via Verizon, but AT&T and Verizon don’t keep detailed track of this traffic and then bill each other. They just figure it will all net out over time.
That approach works fine between carriers that are roughly the same size and pass roughly the same amount of data back and forth. But in the past, Netflix has connected its servers to the internet backbone via several middle tier carriers, most notably one called Cogent. This means that Cogent dumps a ton of data onto the internet, thanks to its contract with Netflix, but since Cogent is otherwise only mid-sized, it doesn’t receive very much traffic in return. Because of this, Verizon and others are increasingly unwilling to simply assume that everything nets out over time. Obviously it doesn’t.
(If you’re wondering why he’s talking about Verizon instead of Comcast, it’s because there’s also a Verizon/Netflix/Cogent dispute. You’ve probably already seen Ben’s piece, but you really should read Kevin’s, too.)
The “you scratch my back, I’ll scratch yours” deal between carriers is clearly a throwback to the days when the internet was a much smaller place, with most traffic running between and among universities and research labs. It’s not surprising to find it breaking down as traffic increases and asymmetries like Netflix (said to be responsible for 30% of all evening internet data in the U.S.) develop.
The problem of paying for my stuff being delivered on your infrastructure was solved long ago by that high tech industry of the 19th century, the railroads. There are many railroads in the U.S., and they own and run the tracks in their domains. To deliver goods from one part of the country to another, they have to work together to hand off goods at their boundaries and arrange scheduling and fair payment. This is the original “last mile” problem.
There are some obvious standardizations that help this cooperation among competitors. The one everyone knows about is track gauge, set at the delightfully weird value of 4′ 8½″.1 But the value itself isn’t important; what’s important is that everyone uses it so the rolling stock can move freely throughout the country. Less well known, but just as essential, are standards for things like knuckles, couplers, and bolsters, which allow equipment to be connected together and replaced if damaged. These standards are the TCP/IP of the Iron Horse.
I don’t claim to be an expert on railroad interoperability, but I’ve gotten some insight by being part of investigations into a few railroad equipment failures. Several years ago I looked into the fracture of a grab handle—one of the bent steel bars used for climbing onto railcars; you’ll see several of them on every car that passes when you’re stopped at a crossing. What I learned was that this handle had failed and been replaced before, probably because it had been whacked by something in passing. Each replacement started with an inspection in a rail yard as the car was being surveyed before being placed in a consist.2 Whenever the handle was found defective, it was replaced by maintenance men for the railroad that owned the yard, and a charge against the owner of the car—another railroad—was generated, and because money changed hands, there was a record of every replacement.
This is all part of interoperability. Each railroad is responsible for the cars on its lines. It gets to inspect and repair the equipment without consulting the equipment owner and then charge the owner for the repairs according to an agreed-upon price schedule. Replace a grab handle: X dollars. Replace an entire wheel set: Y dollars. It’s all covered by a set of interchange rules.
Analogies like this are never perfect and are easy to extend beyond applicability. But if railroads were able to work out a set of mutually beneficial rules for handling each other’s traffic, internet carriers should be able to as well. Packets never have to be taken into the shop for weld repairs.
My maternal grandfather was a track maintenance foreman for the EJ&E Railway—known familiarly as “The J” and recently absorbed by Canadian National. He always claimed that 4′ 8½″ was chosen because it was the distance between Roman chariot wheels, a popular story. This is a prime example of the romance of the railroads. ↩
Consist, emphasis on the first syllable, is a term of art I’m throwing in here to make it sound like I’m an experienced railroad man. Don’t be fooled. ↩