Ranking Modes
The AS rank CGI script provides two methods of AS ranking: degree-based and AS relationship-based modes.
The degree-based mode operates on AS links derived from skitter data. skitter is a tool that uses traceroutes to collect IP traces in the global Internet. To obtain AS links from skitter traces, we map the observed IP addresses to prefixes and then to origin ASes using BGP table dumps from RouteViews taken around the same time as the skitter traces being converted. Then we rank ASes by their degrees (numbers of incident links) in the AS graph obtained as described above.
In AS relationship-based mode, the underlying AS-level topology is obtained from 15 RouteViews BGP table dumps (MRT format RIBs) covering 5 days in 8-hour intervals. Rankings are based on a combination of inferred AS relationships and AS degrees, as described in the general introduction.
Options Available in both Modes
Options Available Only in AS Relationship-based Mode
| dataset: |
Selects the date for which the AS ranking is computed.
The date in the list is the beginning of the 5-day period covered by
a given dataset. The alpha parameter shown is the external
parameter value used to fine tune the multi-objective optimization
algorithm that infers AS relationships. The checkbox labeled prune customer cone using inferred peer-to-peer links (checked by default) determines the precise method of calculating customer cones, which in turn affects the rankings. When this checkbox is not checked, customer cones are computed solely from inferred customer-provider relationships; that is, all AS links are treated as being customer-provider links. In reality, not all AS links are customer-provider links--some are sibling links, others are peer-to-peer links. Thus, relying solely on inferred customer-provider relationships can lead to unrealistic customer cones and inaccurate rankings, though some of the inaccuracy may be mitigated by a good choice of the alpha parameter (we currently apply an automated heuristic to choose the "best" alpha parameter). In contrast, when this checkbox is checked, we allow AS links to be either customer-provider or peer-to-peer, which hopefully leads to more realistic customer cones and better rankings. Peer-to-peer links are inferred from the same BGP paths used in the inference of provider-customer links, though the two types of inferences use differing algorithms. |
| sort ASes by: |
Selects the sorting keys, which are applied in the order
specified, from left to right. The AS ranking is determined
by this sorting. The available keys are
/24s - the number of unique /24 (equivalent) prefixes in the IP address space covered by the customer cone,
prefixes - the number of unique prefixes announced by all ASes in the customer cone, ASes - the number of ASes in the customer cone, and degree - the degree (sum of indegree and outdegree) of an AS. The customer cone is the set of ASes that can be reached from a given AS by following only customer-provider links to both immediate customers and to all indirect customers downstream. |
Options Available Only in Degree-based Mode
| time period: |
Selects the time period of the data to analyze when computing
the AS ranking. The time period is specified with three parameters:
a duration (1 week, 2 weeks, or a month),
an offset (beginning on, ending on, or centered on),
and a base date. Suppose d is the duration (in days) and x is the base date. Then for the offset
beginning on: the time period is x to x+d-1,
ending on: the time period is x-d+1 to x, and centered on: the time period is x-d/2 to x+d/2 (approximately). There are gaps in the skitter AS links dataset. Click on Show Available Dates to see a table of the actual days on which data is available. For the purposes of the AS rank calculations, skitter data is not required to exist for the full time period selected by the user--a certain amount of incompleteness is allowed. Specifically, when the 1 week duration is selected, data must be available for at least 5 days in the selected week; similarly, at least 10 days of data must exist for the 2 weeks duration, and between 20 to 22 days of data must exist for the 1 month duration, depending on the precise length of the selected month. |
Support for this work was provided by the Cisco University Research program and by NSF CNS-0434996, with previous support from DHS/NCS.
![[CAIDA - Cooperative Association for Internet Data Analysis logo]](/images/caida_globe_faded.png)