Internet Access: from Low to High Income Countries

The World Bank report 2021 ‘Data for Better Lives’ focuses on how data can be an asset for all countries and to come to a more equal share of the benefits of data. This is especially relevant for the developing countries. It is a very interesting report. Many topics considering data are dealt with, e.g., enablers of economic progress, safeguards for privacy and security, multi-sided platforms, tax regulation for data traffic and many more.

See report page 308 on infrastucture policies as one of the pillars of an integrated national data system:

The establishment of internet exchange points (IXPs), which requires a competitive market for internet providers, helps create a vibrant digital ecosystem.
Well-functioning IXPs attract content providers locally and from abroad.

One important way to realise more profit from data access in the developing countries is getting affordable internet access. This is an asset to come to a better quality of life. However, there is still quite a mismatch in pricing:

The per-month-costs in the highest category are $140 per MB for broadband and $5.60 per GB for wireless. Compare this to the lowest cost category with only $4.30 for broadband and $0.80 for wireless. Especially, and unfortunately, the developing countries are in the more expensive categories. One might say: ‘poor people pay more for access’. 

One of the reasons is the relatively lack of Internet Exchange Points (IXPs) and ‘ramp-up’ connections (e.g., from cloud providers) to these IXPs in these countries. These IXP connections keep traffic local (‘cached’) and avoid/minimize using expensive lines to get all data from abroad.

See report page 12:

For the most part, low- and middle income countries lack domestic facilities to allow
their own locally generated data to be exchanged (via internet exchange points, IXPs), stored (at colocation data centers), and processed (on cloud platforms)
— Instead, many continue to depend on overseas facilities, requiring them to transfer large volumes of data in and out of the country
— For which they pay a substantial penalty in terms of slower speed and higher prices.

The differences in IXP/ASN connections (ASN refers to the connected organizations connected to an IXP) and centrality metrics will give an indicator of the status of these local exchange, store and process capabilities.

Key in this overview is that we are going to compare these connections for the different income groups as defined by the World Bank and see if we can get some insights. This can help setting up optimal policies.

In my own layman’s terms: the Internet Exchange Point (IXP) reduces the costs of sharing data between the Autonomous Systems Numbers (ASN). IXPs and ASNs together will give an indication of the shared data locally and can be seen as an indicator for level of development. The centrality measures on the ‘bipartite’ network of IXPs and connected ASNs can give a ranking and overview per income group. 

Goal

To come to a view on the worldwide internet development and see the differences between the various income groups:

• Give an overview locations of IXPs and connection of the ASNs to these IXPs
• Set up a graph for this ‘bipartite network’ and use the centrality metrics*
• Identify the status at developing countries from these metrics 

*) A graph and a bipartite network are data science tools to come to an understanding of the important nodes in a network that come from two – therefore ‘bi’ – categories. In this case, the bipartite network consists of the IXPs on the one hand and the ASNs on the other.

Internet traffic and usage per country/income country group are not easy to get since there is not a hierarchical internet organization where you can request this kind of information. 

The above solution using IXPs/ASNs in a ‘bipartite’ network can avoid this and still can give insights that we are looking for: how and where are developing countries progressing? How big is the gap? 

Worldwide maps

A world map with the 4 different income groups and the city of the IXPs plotted as yellow dots:

The countries that do not have a color yet (mostly in Africa) do not have any IXP at all*.

From the above, one may conclude that some lower- and middle-income countries are doing it not so bad at all. Plotting the number of ASN connections per country gives a different picture:

IXPs with large ASN connections can be found mainly in Europe, Brazil, US, Australia, South Africa. India is an exception (being a lower-middle income country, although it also a large country with quite some IT/ICT business). 

*) Notes: 

• IXPs with no ASN connections have been removed from the list and are not shown on the maps.
• The maps are generated with Tableau and 51 cities could not be mapped because these are not uniquely identified. These are not shown on the first map (needs some extra data cleaning).

Top 10 IXPs

Worldwide, in total 424 IXPs are connected to 14,181 ASNs.

The number 1 IXP city with respect to the total number of connected ASNs is São Paulo. This IXP is quite a big one as it has more than twice as much as the number 2 Frankfurt am Main. See above figure, all Top 10 ISP are located in high income or upper middle income countries.

IXPs centrality in ASNs network

Metrics

The used (data science) centrality metrics, to determine the importance of the IXP-nodes, are the following:

• Degree: measures the number of edges connected to the node
• Betweenness: measures the extent to which a node lies on paths between other nodes
• Eigenvector: considers the importance of the other nodes to which it is connected
• Closeness: measures the mean distance the mean distance between a node and other nodes; this is giving a good indication of how far data needs to be retrieved (how many steps)

Overview over all types

The average centrality metrics per income group of the IXPs measured over the whole network for all traffic worldwide:

Income group centrality metrics for type Content with 424 IXPs and 14181 ASNs.

Overall, there is quite a gap at the low-income and lower-middle countries. Interesting to see is that betweenness and eigenvector centrality are higher in the segment upper-middle compared to the high income group. The Brazil-factor (São Paulo) might be a good cause for this. The closeness centrality metric shows that all groups are not that far apart which is good to see for the developing countries: the ‘internet-basics’ are working well: in general, it doesn’t take many steps to get data from one point to another. 

This data can also be plotted per country. As an example, the centrality metrics:

The darker the green, the more important the country’s IXP/IXPs is/are with repect to ASN connections and in relation to all other IXPs worldwide. More countries are not colored at all because they fulfill in international data connections (in the network of IXPs and ASNs) nearly no role.

Overview per subtype

Let’s have a deeper dive into three ASN types: ISPs, Content and Educational/Research.

Income group centrality metrics for type Cable/DSP/ISP with 370 IXPs and 4975 ASNs:

SPs (Cable/DSL/ISP) give a real bad picture. Reason is that especially broadband is lagging in the developing countries due to very high investments needed. Wireless is taking off faster. However, this is a crucial point of difference between the groups.

Income group centrality metrics for type Content with 290 IXPs and 1100 ASNs:

The type Content is unfortunately showing a major difference and lagging behind for the developing countries. Especially, this kind of access to content can help improve the economy.

Income group centrality metrics for type Educational/Research with 271 IXPs and 466 ASNs.

Last but not least, Education/Research gives a winner for the low-income group in the category eigenvector centrality. Apparently, important nodes for this area are already close together. This is also seen in the closeness metrics.

Conclusion

The centrality metrics, as used on a bipartite network of IXPs and connected ASNs, can give a good insight on the differences of the four income groups. In this way, it can be quite easy to compare the different groups and see the status of overall progress. with respect to these metrics:

• In general, low and lower middle income groups have quite some distance to upper middle and high income group.
• The specification per ASN type gives a more detailed picture on which focus would be benificial for developing countries:
  • Recommendation is to focus on promoting to get ASNs local active in the area of Content; this would keep content more local and would make this less expensive for the developing countries;
  • Education/Research is already an area where the developing countries are having data locally available. This is a very positive point.
• The closeness centrality metrics show that the basics of the internet network is altogether good. Developing countries with an IXP could install a proactive policy on attracting ASNs. This could make a jump start in lowering the internet access prices. Of course, countries with no IXP, like in major parts of Africa, should first set up an IXP.

The overall access pricing will lower when more specific ASNs are co-located. It would be even better if ISPs could differentiate in pricing on the type of data.

For further analysis, it would be very interesting to see how the progress in time is developing. Are the low and lower income countries progressing? Can we see the effect of measurements taken? A good thing would be to compare these metrics for the different income groups, and per country, on a quarterly basis.

Affordable internet access is an important driver to come to a better quality of life. It would be very benificial for developing countries to know where to put their focus and limited resources on. The above metrics (e.g. being part of the integrated national data system) can help with setting up optimal policies. Data for good!

See report page 322 (last sentence of final chapter):

And, not least, it is the considered aim of this Report to foster a global discussion that can truly help data improve lives.

Sources and References

World Bank: WDR2021 report.

edX course: Data for Better Lives: A New Social Contract.

Worldbank databank: list of countries classified to income levels.

IXPDB: list of IXPs and related ASNs; see the license for data use.

The data science work has been done in Python by the author.

The three world maps are visualised in Tableau.