The graphs used as a reference point when making the comparisons are built from the BGP information offered by the RIPE RIS and RouteViews projects.
The local routing information added in order to enhance the topology diagrams is BGP data obtained from different Looking Glasses in the Latin America and the Caribbean (LAC) region (CABASE (Argentina), PTT Metro (Brazil), etc.) and from show ip bgp outputs provided by some operators in the region.
The lack of vantage points in Latin America has led to some projects, such as PladMeD, to develop their own platform to gather data from this region.
PladMeD, which is a traceroute-based platform to study the improvements of the first Bolivian IXP, provides us detailed information about Bolivian ASes ecosystem.
For the graphs at the country level, Bolivia is used as a case of study and apart from adding to the national graph the local routing information mentioned above, we also added relationships inferred from an AS paths' set from PladMed.
Relationships between ASes are inferred using CAIDA's AS Relationship inference algorithm.
Although the main focus of this algorithm is to identify the type of relationship between the ASes (Transit or Peering), the type of relationship is not used for this work, just the fact that a relationship between two ASes exists.
It is important to note that mainly two different criteria can be used to restrict an Internet topology graph to a certain area: 1) use the information of the country where the organisation to which the AS was assigned is based; 2) use geolocation information in order to determine the countries in which an AS is active, i.e., the countries to which the prefixes being announced by an AS are geolocated.
Taking into account that an AS is not necessarily used in the country to which it was assigned, the first option is not used.
Instead, the prefixes announced by all the active ASes were geolocated using RIPEstat Data API in order to determine the countries where each AS is active.
To create the national graphs for Bolivia and the regional graphs for the LAC region, we filtered the World graphs in order to include just the ASes that are active in the area of interest and the relationships in which these ASes are involved.
Although criterion 2) is more accurate than 1), there are some misleading entries on every geolocation database.
In this case, RIPEstat shows that AS701, AS1239 and AS10434 are active in Bolivia but no dataset has shown a link between these ASes and other ASes truly participant in the Bolivian network.
Due to this mistake, Bolivian graphs are not completely connected.
Looking for comparing the graphs we used the following parameters: |V| (number of vertices), |E| (number of edges), ⟨d⟩ (average degree), max(d) (maximum degree), ⟨cc⟩ (average clustering coefficient), ⟨k⟩ (average shell index) and max(k) (maximum k-core).
The average degree is computed as ⟨d⟩=2|E|/|V| and it measures the number of relationships in which an AS is involved on average.
The average clustering coefficient as 1/|V|Σ 1 ≤ i ≤ |V| cc(i), where cc(i)=(m(i))/(d(i)(d(i)-1)) is the clustering coefficient of vertex i, d(i) is the degree of vertex i and m(i) is the number of edges between the neighbours of i.
The cc measures the level of interconnection within a node's neighbourhood.
The k-core is a subgraph where all vertices have at least degree k.
The k-shell-index is the maximum core a vertex belongs to.