Steve, To summarize why folks want to do this: Most of the really big ISPs peer at the OC12 or OC48 level with each other. The provisioning interval on these large circuits is insane. There is typically little or no cost - it's not really about expense, as the Big ISPs are also big carriers, and the do meet in certain places, circuit-wise. The problem is that the carrier arms of these companies move slowly, especially for such large ciruits. They may also through up political roadblocks. And, in those occasional cases where a large ISP needs to get a circuit to reach another carrier, that OC-48 really hurts financially. These ISPs want to meet in facilities that have room and power for all their transport gear and routers. They want stuff like multiple entrance facilities, high security, and other features. They want to be able to just run x-connections to each other when they need a new circuit. Of course, it's not quite so simple, because if someone is running an OC-48 x-connect between two really big providers theirs lots of other work to do, but its still a real time-saver. I don't think anyone disputes the utility or engineering good-sense of this proposal - I certainly do not. However, there seems to be an assumption amongst some posters, that just because something makes good sense from an engineering point of view, then it's a good thing. This could be a great thing for the evolution of the Internet. It could even help the small ISP, by saving significantly on peering costs, by identifying certain sites as private interconnection sites. On the other hand, it could lead to higher pricing, a willingness by the larger providers to work more closely together, and a more exclusionary attitude towards peering. These would be bad. There's also an engineering problem with this approach. When folk prodominantly peered at the public exchanges, they were point-of-failure, and a MAE going down could be felt rather acutely across many networks. By peering in the same facilities, there many be greater risk exposure for the Internet as a whole. Sure, there's less risk of individual circuits going down, but greater risk of longer impairment to routing due to fire, loss of electrical power, or flooding in one of these sites. It's the risk of centralization vs distribution. Just thought this was worth pointing out. (And no, none of these connections will go through a central switch, it's private peering. That does reduce the risk, which is why I do not mention equipment failure as a possible risk) - Dan
-----Original Message----- From: owner-nanog@merit.edu [mailto:owner-nanog@merit.edu]On Behalf Of Steve Feldman Sent: Wednesday, July 25, 2001 9:29 PM To: nanog@merit.edu Subject: Re: The large ISPs and Peering
Ok, my turn to weigh in on this. :)
First, can we stipulate that the "big players" have to peer somewhere?
As I see it, there are two extremes:
1) a mesh of point-to-point circuits between sites, (SONET, dark fiber, metro-gigE, etc.)
2) common colocation with local private links between routers
On the whole, option 2 seems easier to manage and potentially less expensive. Here's why:
- There can be problems getting circuits between carriers. I once heard from an engineer at one carrier-owned ISP that a private peering circuit to another carrier-owned ISP had been on order for over a year, because each carriers refused to allow the other to carry the traffic on its fiber.
- It should cost less to bring a few large circuits into a central facility and use the colocated router to multiplex the cheap local peering circuits, than to provision a bunch of smaller metro circuits.
- It's much quicker to run fiber across a room than get a circuit from any Telco. This makes additions and changes much easier.
Sure, a carrier-owned ISP likely can get ciruits into the colo more easily, but it's not that expensive any more for others to get in. There are a bunch of new companies building carrier infrastructure and trying to sell cheap metro high-speed circuits into popular buildings (I worked for one until recently), and dark fiber is often available for even less.
As for the political question of peering policies, I won't get into that except to say that it seems an orthogonal issue to me.
Just my $.02, Steve