AS number related metrics

AS number related metrics

11 December 1997, hwb/kc -- NLANR and CAIDA


some problems are just too complicated for rational logical solutions.
they admit of insights, not answers.

jerome bert weisner, A Scientist's Advice II, New Yorker 26 jan 1963


Autonomous system (AS) numbers in BGP-based routing tables are, with some degree of risk, indicative of how networks and ISPs are interconnected. Those BGP tables offer an AS-based path from the table-generating BGP peer to a target network/mask pair, i.e., a destination network. Furthermore in some cases it may be useful to consider via which AS a specific network was learned, though the degree of risk is even higher here, as the routing table may indicate that the information is incomplete.

While it would be useful to map AS numbers to ISP, many ISPs hoard many AS numbers, and there is no data base that one could ask simple questions such as "tell me all the AS numbers belonging to a specific ISP." This is further complicated by current and pending mergers and restructuring of the ISP landscape. As an example, one of the most successful collector of ISPs appears to be Worldcom. Trying out the hard way hop many AS numbers belong or will belong to Worldcom, one has to ask and look around. A possible answer may be:

In other words, abstracting AS numbers to ISPs is not a straightforward undertaking. But then, neither is abstracting multiple network prefixes to AS numbers, as there is no one-to-one relationship, and also networks are at times connected via multiple ISPs.

Hence, while there are no really good mappings that would accommodate this in a straightforward and "absolutely correct" sense, when allowing for a margin of error, such abstractions may still yield valuable insight into the structure of the Internet system. Insight that would be hard to get otherwise. Therefore BGP tables from Internet routers or measurement systems (e.g., the one described in http://www.antc.uoregon.edu/route-views) can be valuable instruments.

Example questions for network metrics to be asked include:

  1. what percentage of all ASes are related to a specific ISP
  2. what percentage of all used (BGP-known) ASes are related to a specific ISP
  3. what percentage of ASes in the BGP routing system have adjacencies related to a specific ISP
  4. what percentage of all AS routing announcements are related to a specific ISP, perhaps based on BGP traffic traces
  5. what percentage of all total address space (via network prefixes) are announced for from a specific ISP
  6. what percentage of all allocated address space (via network prefixes) are announced for from a specific ISP
  7. what percentage of all announced address space (via network prefixes) are announced for from a specific ISP
  8. what percentage of all traffic does a specific ISP route to, perhaps based on packet traces
  9. what percentage of all Internet transit-related revenues is a specific ISP responsible for

given the aforementioned Worldcom example, the answer to question (2.) would be for the 10 December 1997 to be 143 out of 3154 known AS numbers. However, in answer to question (1.), the InterNIC's 26-Aug-97 asn.txt file (minus the reserved space) reports 7383 assigned AS numbers, out of the 2^16 or 65536 total, i.e., many more AS numbers are assigned than actually known to the global interconnected BGP system.

The following graph describes the number of BGP adjacencies based on the BGP information, and considers the subset that neighbors the above mentioned Worldcom related AS numbers, with the lower graph describing the Worldcom fraction.



Going a step further, one can assess what address space is being advertised, what address space is injected to the BGP system via specific clusters of AS numbers (ignoring for the moment whether the information was learned from an IGP or derived from incomplete information), and what address space is no more than one AS hop away from a specific AS (again, ignoring for the moment whether the information was learned from an IGP or derived from incomplete information). This is illustrated in the following graph.