Kevin,
As evident in the stories people have posted, we get at least one major peering dispute per year between tier 1 providers and certain pockets are cut off from other pockets. Mind you that it's not everything cut off for those people; just certain destinations. This is the nature of peering disputes and it results in some temporary breaks in the IP infrastructure. The fact that you can route around it above the IP layer of the OSI model with proxy or DNS tricks is useful to a minority, but not to the majority that is being affected. BGP propagation is delayed (or dampened) by design. You do NOT want a flapping interface on a single router on the Internet that's going up and down every few seconds or few minutes to cause a massive ripple of BGP updates throughout the Internet every few seconds or minutes. For this reason, BGP updates are very conservative for good reason and they are designed to ignore temporary failures. Only when the failure is prolonged will the BGP routing update be sent out to the Internet. When the original link comes back online, there is once again a delay on the BGP update because we can't be sure if the link is just flapping up and down or if it's permanently back online. As for the concept of fault tolerance on the Internet, that is a capability of certain portions of the Internet and it must be carefully designed, engineered, and it is a premium feature. Failover is not a default design on the Internet nor should anyone expect it. The concept of auto failure or this massive meshed network on the Internet is largely a myth. Every failover route is by careful design and someone is paying handsomely for that capability. George From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Kevin McArthur Sent: Friday, October 31, 2008 5:11 PM To: George Ou Cc: 'NNSquad' Subject: [ NNSquad ] Re: Peering dispute cuts off Sprint<->Cogent Internet traffic George, In this case however, we're talking major carriers, which have many many peers (massively multihomed). There are multiple routes around, as evidenced by the fact that people are using proxying services to get around the damage at the application layer. As for BGP propagation, if its following the RFC then the routes are already advertised and before the peering agreement broke down, they just happened to be a shorter path. Once that peer dies, it should fall back to the longer paths in near real-time. (within moments, as soon as the bad route stops advertising (which it cant do if it cant reach the other end)) This is a core routing principle of the internet, and how it is supposed to be tolerant to attacks on infrastructure -- If this mechanism isn't working, then we have some serious resiliency problems on critical backbones. You would have to do something special to stop BGP from rerouting -- like, for example, falsely advertising a route that doesn't work while making it appear closer than the alternatives. You could also theoretically block the remaining peers from advertising routes to that network, but again, that would be a massive net neutrality violation as they would be actively blocking a pathway, and not simply just not peering. Essentially, saying if I don't want to peer with you, no one else can either. There's more going on here.... Kevin McArthur George Ou wrote: There's no violation of any RFCs here, it's a peering dispute which is quite common on the Internet. It's a long running myth that routes are automatically rerouted on the Internet. Unless one of the two end-points is dual-homed with 2 completely separate ISPs configured for BGP (or DNS remapping), any break in the route means a disconnection between the two points. Even when BGP does exist, it takes some time for the routes to propagate so there's always some outage for a period of time when there's a break in the link. George -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Barry Gold Sent: Friday, October 31, 2008 1:38 PM To: NNSquad Subject: [ NNSquad ] Re: Peering dispute cuts off Sprint<->Cogent Internet traffic From: Ed Jankiewicz <mailto:[EMAIL PROTECTED]> <[EMAIL PROTECTED]> Subject: Total Filtering As many news organizations are now reporting, Sprint-Nextel (Embarq) has decided to sever its Internet connection with Cogent, another Internet service provider. This action has caused a "hole" or "rip" in the internet, meaning that Sprint-Nextel (Embarq) and Cogent customers may find they cannot access resources hosted by the other company's customers. Similar standoffs have occurred in the past, and usually one company backs down after a few days, but no one can predict what will happen in this case. OK, so what has happened to the "treats censorship as damage and routes around it" Internet? Even if Embarq and Cogent are no longer talking to each other, the routers should be automatically finding routes via other carriers and sending the packets -- around Robin Hood's barn if necessary, but the Internet is supposed to be _robust_. Jon Postel designed it that way -- I've read the RFCs. That's what ARPA specified when they paid for the development of first the ARPANet and later the Internet -- and what NSF paid for when they branched off NSFNet and allowed commercial traffic. Are these guys programming their routers to just drop packets with certain destination IP addresses, instead of finding the shortest available route? I'm beginning to think that Congress (or perhaps an international body similar to the WTO) should make the core RFCs (IP, TCP, BGP, FTP, HTTP, SMTP, and RFC 822) have the force of law. And anybody who violates those protocols should be fined and/or have their connections turned off. -- Kevin McArthur StormTide Digital Studios Inc. Author of the recently published book, "Pro PHP" http://www.stormtide.ca
