On Thu, Oct 06, 2005 at 10:54:37PM -0700, JC Dill wrote:
Various people have stated that uneven data flows (e.g. from mostly-content networks to mostly-eyeball networks) is a good reason to not peer. I'd love to know how it improves Level 3's network to have data from Cogent arrive over some *other* connection rather than directly from a peering connection. Do they suddenly, magically, no longer have backhaul that mostly-content data across their own backbone to their users who have requested it if it should come in from one of their *other* peers who (in normal peering fashion) hot-potato hands it off to them at the first opportunity, rather than coming in directly from Cogent?
I don't think so.
So why break off peering???
AFAICT there's only one reason to break off peering, and it's to force Cogent to pay (anyone) to transit the data. Why does L3 care if Cogent sends the data for free via peering, or pays someone ELSE to transit the data?
First off, why do you assume that peering is a right to which people are entitled? Level 3 operates a network, a pretty darn big and successful one (well big at any rate), and it apparently does a good job delivering the bits or it wouldn't have as many customers as it does. Why *must* they give another network free access to their network if they don't feel that it is mutually beneficial? You seem to be making the argument that because YOU think it is mutually beneficial, therefore (3) must be doing something evil. Beneficial is in the eye of the beholder, and Cogent could just as easily have been the one to decide it wasn't beneficial to them. Second, there are serious some serious fallacies in the argument that "the bits have to go there anyways". The vast majority of the Internet is multihomed in one way or another, especially the closer you get to the big Tier 1's. By my count, Level 3 has over 57,000 customer prefixes. I don't have any data about how many of those Cogent was actually using before-hand (though I'm sure someone does), but we know that roughly 4,000 some prefixes were single homed and couldn't be reached any other way. I'd be willing to venture a guess that while Cogent was probably not using anywhere near 57,000 prefixes from (3), they were using a heck of a lot more than 4,000 before the depeering notice and traffic depref, and that those prefixes were substantially trafficked. There is also this thing called BGP path selection, and one of the important criteria is AS-PATH length, so making the path to 3356 longer may have diverted a significant amount of traffic away from Cogent at the source of their multihomed customers. A depeering would combine both effects, Cogent would immediately depref, and the longer AS-PATH would shift traffic away from this path. Even if reaching (3) didn't cost Cogent a dime, depeering may still have been a viable method of traffic engineering. The bottom line is that we have no idea what was really going on, but there are loads of reasons why (3) would want to depeer Cogent that don't have anything to do with forcing them to pay for transit. I don't think anyone who depeers another network actually expects to see a dime in transit business from the depeered network any time in the immediate future anyways.
I think this is about a big bully trying to force a smaller player off of the big guys' playing field (tier 1 peering). From where I sit it looks like an anti-competitive move that is not a "best effort" to serve their customers but a specific effort to put another (smaller) competitor out of business (of being a transit-free or mostly transit-free backbone) by forcing them to pay (someone), forcing their costs up. Level 3 must know they are no longer putting for a "best effort" for their own customers to connect them to the "internet" (as their customers see it, the "complete internet" that their customers have come to expect).
If Cogent can't stay in business if (3) decides that there is no benefit to giving them free access to all of their customers, and infrastructure necessary to deliver it, they were never a peer to begin with. Peering is about mutual benefit, not entitlement. Cogent makes the same value judgements when it decides if it is going to peer with another network or not, it is no different here than it is there. Cogent routinely turns away smaller peers for peering and suggests that they buy transit instead, are you going to accuse them of anti-competetive practices next?
I also believe that Cogent has a valid argument that Level 3's behavior is anti-competitive in a market where the tier 1 networks *collectively* have a 100% complete monopoly on the business of offering transit-free backbone internet services. As such, L3's behavior might fall into anti-trust territory - because if Cogent caves in over this and buys transit for the traffic destined for L3 then what's to stop the rest of the tier 1 guys from following suit and forcing Cogent to buy transit to get to *all* tier 1 networks? Then who will they (TINT) force out next?
What exactly is the business of selling transit-free backbone internet services? Being a transit free backbone is a silly marketing ploy at best, at worst it makes you inflexible and vulnerable to exactly the kind of outages (3) is suffering when someone refuses to transit routes to the. Besides, I don't think (3) or any of the other large networks are doing all that well financially with their monopoly. If anything this makes some strong points about exactly how non- monopolistic the situation really is. Think about it, despite the fact that Level 3 provides transit to 1/3rd of the Internet by prefix count (a silly measurement I realize, but its all I've got :P), the vast majority of the customers and the Internet manage to route around this particular issue. That is because the vast majority (over 90% it seems) of (3)'s customers are multihomed, and those networks can be routed around or donut peered around. For example, several large cable MSOs are (3) customers, but peer with other networks. My company peer with a lot of them, and according to our netflow statistics we donut around over 95% of the potential traffic that could go to a (3) peer by peering with their customers and alternate providers to multihomed customers. This is the kind of thing that is happening across the world and across the board, most sizable networks are perfectly capable of bypassing the Tier 1's for the vast majority of the traffic. In fact that is one of the key reasons why Cogent, a much smaller company, is as large or larger than (3) in terms of absolute traffic volume. The Tier 1's are in a losing battle, as more and more of the bits that they used to handle and bill as on-net traffic is shifted away from them, and instead move directly between major content networks and major eyeball networks. It is not unexpected to see them tightening policies around the customers that they DO have, it is one of the few resources they have.
What's to stop a big government (like the US) from stepping in and attempting to regulate peering agreements, using the argument that internet access is too important to allow individual networks to bully other networks out of the market - at the expense of customers - and ultimately resulting in less competition and higher rates? Is this type of regulation good for the internet? OTOH is market consolidation good for the internet?
I don't like this slippery slope, I don't like it one little bit.
Hopefully, sanity. Whether you believe it or not, the Internet *IS* best effort. If you don't like it, or you aren't happy with it, buy from someone else who makes a better effort. Speaking of slippery slopes, how would you like to not be able to block traffic to your network from people you consider to be spam sites, because you are harming global reachability and potentially trying to force those sites out of business? Same argument, different benefit for you. -- Richard A Steenbergen <ras@e-gerbil.net> http://www.e-gerbil.net/ras GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)