These are the signs of the apocalypse: A worldwide earthquake, the sun ceasing to emit visible light, cats and dogs living together in harmony, and Cogent (AS174) depeering another AS. At least one of these happened earlier this week.
At about 10:00 UTC on Tuesday (6am EDT), Cogent depeered a couple of smaller, UK-based ISPs without notice. This was apparently intentional and due to a review of existing peers and whether they meet peering policies. Does this mean that Cogent is becoming more like its larger competitors that it so enjoys taunting? I’ll take a look at who was depeered and speculate on why.
This is the first in a two-part post about what happened to Cogent on Tuesday, April 24, 2007. Later that day, Cogent suffered what appeared to be a fairly widespread serious routing problem. Looking at that in a bit more detail will be a subject for part 2.
Some background is useful. To use a turn of phrase from the former peering manager of a large European network, Cogent is one of the most massively “overpeered” networks in the world. They sit at the top of the Renesys Market Intelligence Peering Base ranking. That ranking is a simple sum of the Customer Base score of a network’s peers. (Readers who are newer to networking may wish to review some of the basics of peering.) Cogent has hundreds of peers (250 is a lower bound and probably closer to 400), and they include many very large networks. They also include many very small networks. Cogent, along with Abovenet, is known for having a very open peering policy driven by a combination of history and opportunistic cost savings.
So who was depeered? The former peers that were disconnected include:
- Telecomplete Ltd (AS6320)
- Entanet Ltd (AS8468)
- Datahop (AS6908)
Datahop was apparently depeered a week ago, but Entanet and Telecomplete both dropped at roughly the same time this week. These three are all UK-based providers, reasonable-sized (top-100 ranked in the UK certainly). Still, this is not really a startling event: huge top-15-in-the-world network depeers a few UK-based networks who were lucky to have ever had peering. Something like that—except for what it might portend.
There are two relevant factors here to muse about: Europe and Cogent’s relative maturity as a company.
Firstly, Cogent has a fairly serious Europe problem right now. They have been aggressively attacking the European market for a few years now and making some solid headway. They bought a couple of carriers (Lambdanet Spain and France, Carrier1 in Germany among them), ruthlessly integrated them and then proceeded to undersell the market by a factor of 50-80%. This has made them many enemies.
As a result of this approach to business, Cogent has much less effective peering in Europe than do many of its larger competitors. Most of the European PTTs refuse to peer with Cogent anywhere on the European continent. Recently, some large US carriers (among them Level (3) ) seem to have adopted a similar approach. This means that when Cogent sells capacity in Europe, it is forced to drag that traffic back to the US to hand it off to its peers here. Of course that means that if the ultimate destination is European, the traffic has to travel back. This is a burden on both Cogent and the European carrier and, of course, the customers on both sides. But it’s unlikely to change because of just how much hate there is for Cogent among European networkers.
Given its Europe problems, it’s a little odd that Cogent would depeer anyone who actually peered with them in Europe. The reason Cogent allegedly gave was that the peerings did not meet their peering policy requiring “pan-European peering”. The only explanation that looks plausible is that Cogent is beginning the long process of grooming its smaller peers as part of an effort to improve its traffic to larger peers and perhaps build a case for new European peerings in the future. It will be interesting to see how that plays out.
But that brings us to the second issue: Cogent’s maturity as a company. If Cogent is pruning small peers as part of such a longer term strategy, that means that they’ve reached the early adulthood of their network. They are no longer freewheeling teenagers willing to hook up with whoever seems fun at the time. They are cutting ties to some of their less-appealing friends as part of a strategy to seem more appealing to future mates. Given the Brits’ reputation for drunken loutishness, it’s hardly any surprise that they were the first to get the axe!
If Cogent is settling down into its corporate middle age, interesting things are afoot. Cogent has long held the role of wacky, freewheeling, young whippersnapper—the new kid on the block willing to undercut everyone’s prices, maybe while cutting a few corners on coverage and network reliability, but not so much so that its customers aren’t happy for the deal. That role has been one of the big drivers pushing IP transit prices down. Someone has to be the “low price leader” and for the past several years, that has been Cogent. If these depeerings are not isolated events, but rather part of a new strategic direction for Cogent, that could spell changes in pricing and competitive dynamics.
It seems far fetched, I’ll admit, but I would have said the same thing about the odds of Cogent depeering anyone (let alone three peers in the same week).
Next time, I’ll look at the backbone outage that happened the very same day as these depeerings and try to come up with some reason why that wasn’t just an odd coincidence.