We study the growth trends in the number of autonomous systems (ASes) seen in the global routing system. We also study the number of ASes according to registry. The data used for this analysis is derived from Routeview and RIPE BGP dumps over the last 12 years. To map ASes to registries, we used Team Cymru's WHOIS lookup service. More information about the data collection and pre-processing is in Amogh Dhamdhere and Constantine Dovrolis' IMC paper, "Ten years in the evolution of the internet ecosystem" and in the paper's supplemental data page.
A comprehensive analysis of AS Number (ASN) trends by Geoff Huston is available online at The 32-bit AS Number Report. That analysis studies AS allocations globally and according to registry over time. Huston also studies advertised vs unadvertised ASNs, but only for the global case. Here, we focus on advertised (as opposed to allocated ASes), and split them according to the registry from which they were allocated.
Our main observation is that the two largest registries in terms of the number of advertised ASes (ARIN and RIPE) have shown distinctly different growth trends since 2001. Both registries showed exponential growth until mid-2001. Since then, ARIN has grown linearly, while RIPE changed to a slower, but still exponential, increase. The number of advertised ASes is now larger in RIPE than ARIN.
The growth trend for all ASes until 2001 is well modeled by an exponential fit, with exponent 0.091. The growth trend changed around mid-2001. Visually, the trend post-2001 appears close to linear. However, an exponential model with a lower exponent (0.03) than in the first stage gives a better fit (better R-square, and smaller sum of squares error).
We study the growth trends in the number of ASes according to registry. ARIN and RIPE, the two biggest registries, have distinctly different trends. Both of them show an exponential increase until mid-2001. Since 2001, however, the number of ASes in RIPE has grown faster than ARIN. The number of ASes in RIPE is currently larger than in ARIN. Visually, it appears that ARIN has grown linearly post-2001, while RIPE has grown exponentially.
Next, we obtain regression fits for each of the registries separately.
The growth trend for ARIN ASes is exponential until mid-2001, with an exponent of 0.085, which is just slightly less than the exponent for all ASes. Thereafter, a linear growth with a slope of 211 ASes/snapshot (a snapshot duration is 3 months) gives the best fit.
The growth trend for RIPE ASes is exponential until mid-2001, with an exponent of 0.095, which is also quite close to that of all ASes. Thereafter, exponential growth with a smaller exponent (0.046) gives a good fit upto 2009. Since 2009, the growth of RIPE has slowed down. As we have just 4 snapshots after 2009, we do not attempt to model the trend after 2009.
The growth trend for APNIC ASes is exponential until mid-2001, with an exponent of 0.097, slightly greater than the exponent for all ASes. Thereafter, an exponential growth with an exponent of 0.03 (similar to the exponent for all ASes post-2001) gives a good fit.
The growth trend for LACNIC ASes is exponential until mid-2001, with
an exponent of 0.084, which is lower than the exponent for all ASes
pre-2001. The growth after 2001 can be reasonably well modeled with a
single exponential (correlation coefficient 0.98). A combination of 3
linear segments gives a better fit, however:
1) mid 2001 to early 2004.
2) early 2004-early 2007.
3) early 2007-present.
The growth trend up to 2001 is not well modeled by either exponential or linear. The exponential model gives a better fit, but the correlation coefficient is only 0.9. The growth after 2001 is well modeled as exponential, with an exponent of 0.056, which is greater than the exponent for all ASes post-2001.