In the context of Internet routing, the default-free zone (DFZ) refers to the collection of all Internet autonomous systems that do not require a default route to route a packet to any destination. Conceptually, DFZ routers have a "complete" BGP table, sometimes referred to as the Internet routing table, global routing table or global BGP table, but, realistically, the widespread use of route filtering and the rapid rate of change in Internet routing ensure that no router anywhere has an absolutely complete view of all routes, and any such routing table would, in any case, look different from the perspective of different routers, even if a stable view could be achieved.
Contents |
The Weekly Routing Reports [1] used by the ISP community come from the Asia-Pacific Network Information Centre (APNIC) router in Tokyo, which is a well-connected router that has as good a view of the Internet as any other single router. For serious routing research, however, routing information will be captured at multiple well-connected sites, including high-traffic ISPs (see the "skitter core") below.
As of September 17, 2011, there were 371690 routes seen by the APNIC router. These came from 38823 autonomous systems, of which only 134 were transit-only and 32200 were stub/origin-only. 5208 autonomous systems provided some level of transit.[2]
The term "default-free zone" is sometimes confused with an "Internet core" or internet backbone, but there has been no true "core" since before the Border Gateway Protocol (BGP) was introduced. In pre-BGP days, when the Exterior Gateway Protocol (EGP) was the exterior routing protocol, it indeed could be assumed there was a single Internet core.
That concept, however, has been obsolete for a long time. At best, today's definition of the Internet core is statistical, with the "skitter core" being some number of AS with the greatest traffic according to the CAIDA skitter measurements[3]. The CAIDA measurements are constantly updated.
Large Internet Exchange Points (IXP)—in that they typically include full routes as seen by multiple ISPs, as well as customer routes, in their exchange fabric—are extremely good places to assess global Internet routing [4].
Before the current commercial Internet evolved, the NSFNET, which interconnected five US government funded supercomputer centers, could have been considered the high-speed Internet core. Four IXPs supported NSFNET, but these IXPs evolved into a model where commercial traffic could meet there. While it is slightly difficult to point to a precise endpoint, NSF funding for transmission ceased by 1998.
It's quite common practice, in a multihomed but stub (i.e., non-transit) autonomous system[5], for the BGP-speaking router(s) to take "full routes" from the various ISPs to which the AS is multihomed. Especially if there is more than one router connected to the same ISP, a common practice, it will receive more routes that are in the DFZ. Here's the reasoning: when you have two routers connected to a major ISP such as Sprint, France Telecom or Qwest, that provider has a number of customer AS connected to it. The optimal route to those customer AS are important to the ISP itself, but also tells one customer AS which specific router has the best path to the other customer. The "full routes", or properly "full routes plus customer routes", coming to a customer router makes that customer router part of the DFZ, but certainly not part of the current concept of the "skitter core".