This lecture by Harry Balakrishnan covers a high level overview of Interdomain Routing in today's internet. Internet is constructed of many Autonomous Systems (ASes) which can be quite different in their internal structure and routing protocol. Each AS is usually owned by a different company and is in competition with other ASes (also referred to as ISPs). What enables the global connectivity over internet is the routing protocol in between ISPs. Current wide-area routing protocol in use is called Border Gateway Protocol (BGP, Version4).
BGP is interestingly different than other routing protocols since its primary goal is global reachability between ASes with special focus on scalability. BGP is centered around following the policies which are primarily affected by the relationship between ASes and ultimately profits. In a usual Interior Gateway Protocol (IGP) the main concern is minimizing some kind of path metrics.
The paper gives a good overview of the process in which different ASes choose to advertise (and therefore claim availability of usage) routes to other ASes and how they absorb the routes advertised by others into their internal structures. The author also discusses hijacking routes by false advertisements, convergence speed and multi-homing as some of the main challenges of today's wide-area routing.
I found the provider-transit vs. peer-peer forms of relationship among ASes quite interesting. Why would peers which are carrying possibly more than 2 times of the traffic possibly worth millions of dollars make such agreement? (well I'm sure there are good reasons for it!) Is the required billing infrastructure too expensive to justify such approach? Furthermore, why ISPs can not have mixed relationships? e.g. ISP A which has two other peers B and C, can act as a provider of B for the traffic targeted to C and act as a peer for the traffic targeted to A and its transits? or maybe this capability already exists and is in use.
At the end I found the lecture a well written overview of BGP protocol.
BGP is interestingly different than other routing protocols since its primary goal is global reachability between ASes with special focus on scalability. BGP is centered around following the policies which are primarily affected by the relationship between ASes and ultimately profits. In a usual Interior Gateway Protocol (IGP) the main concern is minimizing some kind of path metrics.
The paper gives a good overview of the process in which different ASes choose to advertise (and therefore claim availability of usage) routes to other ASes and how they absorb the routes advertised by others into their internal structures. The author also discusses hijacking routes by false advertisements, convergence speed and multi-homing as some of the main challenges of today's wide-area routing.
I found the provider-transit vs. peer-peer forms of relationship among ASes quite interesting. Why would peers which are carrying possibly more than 2 times of the traffic possibly worth millions of dollars make such agreement? (well I'm sure there are good reasons for it!) Is the required billing infrastructure too expensive to justify such approach? Furthermore, why ISPs can not have mixed relationships? e.g. ISP A which has two other peers B and C, can act as a provider of B for the traffic targeted to C and act as a peer for the traffic targeted to A and its transits? or maybe this capability already exists and is in use.
At the end I found the lecture a well written overview of BGP protocol.
In general, peering relationships are historical. Big ISPs do not peer with small ISPs, only similarly large ones, although they may inherit such relationship through business acquisitions. Once you become a customer, you never become a peer later.
ReplyDelete