2025 Report

Each section below describes the results attained by analysing all the gathered BGP routing data.

Click on the items in the legend will show/hide the items in the plot.

ASN Visibility

These plots show how many ASNs each network has visibility of, and what percentage of that visibility overlaps with other networks:

  • Visibility of All ASNs Seen in The DFZ - This plot shows how many ASNs were visible to each network (and what percentage that is of all ASNs that were seen across all the gathered BGP data). I haven’t had time to look into why Cogent and Angola Cable have such a dip in the number of ASNs they see, when compared to how many are in their reconstructed tables. I am expecting this is the bogon ASN, bogon prefix, and prefix length filters kicking in. I couldn’t reproduce a full table for Virtual Technologies so they have a valid excuse.

  • % of Visible ASNs (Not-)Visible via Other Networks - For each network, this plot show what percentage of the ASNs that network has visibility of, are visible by at least one other network. For Tier 1s this really should be as close to 100% as possible, across the board; they should all have full visibility of all ASNs in the DFZ because that is what they are selling, full DFZ connectivity, and because they should have a full mesh of peerings between each other. The small percentage of ASNs visible to only a single network are presumably due to either technical limitation (e.g., the NO_EXPORT community being used, or one company operating multiple ASNs internally), or geopolitical or regulatory requirements. Non-Tier 1s doesn’t need to have visibility of 100% of ASNs; it’s possible to reach 100% of prefixes without having reachability to 100% of ASNs and hosts need reachability to IPs/Prefixes not ASNs.

  • % of ASN Visibility Overlap - This plot shows for any pair of networks, what percentage of the ASNs they have visibility of, overlap between those two networks. The lower the percentage, the more disparate the set of ASNs each network has visibility of. Also note that because two networks each have visibility of a different amount of ASNs, the percentage of their overlap differs based on their two perspectives. 78k ASNs are visible to both Verizon and Cogent, but that is 98.98% of ASNs that Cogent’s sees, but only 94.6% of ASNs that Verizon sees, because Verizon sees more ASNs than Cogent, and because they see a different set of ASNs. This means the percentage value is different if you are looking at the heat map horizontally or vertically.

Direct Peerings

These plots show how many directly connected networks each network has, where that connectivity is, and what impact that has on average path length:

  • % of All ASNs Seen in The DFZ Which Are Directly Connected - This plot shows the how many of all the ASNs which were seen in the DFZ, each network is directly connected to (the first hop in a BGP route). It’s not surprising that the networks with the least number of direct peers are also incumbent national telcos which claim to be Tier 1s. Whereas the the networks with the most directly connected peers are known for either being a Tier 1 with a focus on providing global connectivity, or a non-Tier 1 which has has put significant effort into their global peering operations and deserve to be where they are.

  • (Non-)Unique Directly Connected ASNs - This plot shows how many directly connected ASNs are directly connected to a single network only. I think it shows the difference in business goals; i3D for example, who have excellent peering, aren’t interested in provide IP transit connectivity, so whilst they are one of the most peered networks in the world, almost none of their peers are unique (their goal is to reduce latency for their gaming customers). Whereas the top 6 networks in this plot are all actively selling IP Transit and thus will have a good number of connectivity customers single homed to them. For some of the networks which aren’t well peered, considering they are Tier 1s e.g., Verizon, DTAG, etc., a significant percentage of their direct peers aren’t directly connected to any of the other networks being examined.

  • % of Overlapping Direct Peers - This plot shows the overlap of directly connected ASNs between pairs of networks. As with the % of ASN Visibility Overlap plots, the number of overlapping ASNs is the same for each pair of networks, but what percentage that is to each network will differ based on the number of peers they each have.

  • Breakdown of Directly Connect ASNs by Continent - This plot shows the breakdown of directly connected ASNs by continent, to show where each network has the strongest geographical BGP presence. It comes as no surprise that the Tier 1 networks who’s origin story is being a national incumbent telco have little connectivity to ASNs outside their home market, whereas those which strive to be global connectivity providers or have extensive global peering have a more balanced BGP geography. Each ASN was checked against the NRO allocation data, this comes with the caveat that an ASN may be registered to a company in country X, but used in country Y. This means the plot is broadly accurate but it can not be 100% accurate because some ASNs are operating in nearly every country in the world.

Hop Count

  • AS Hop Count Frequency (<= 10 hops) - This plot shows the number of ASNs each network can reach in <= 10 hops (only a tiny percentage of ASNs are not reachable within 10 AS hops). Here 1 hop means directly connected. If an ASN was seen via multiple paths of different lengths (it was reachable in 1 hop because it is directly connected, and in 2 hops via a peer), only the shortest path is counted.

  • Weighted AS Hop Count (To ASNs <= 10 Hops Away) - Most networks aren’t reachable in 1 hop and the lower the hop count, the more control a network has over the path to the destination. This means that a weighted average is needed. 10 points are assigned for each ASN which is reached in 1 hop, 9 points for each ASN reachable in 2 hops, 7 points for 3 hops, and so on. This plot shows the weighted average. This is why RETN ranks better than Convergence even though the biggest group of ASNs reachable via Convergence are reachable in 2 hops, and the biggest group of ASNs reachable via RETN are reachable in 3 hops, Convergence has a longer long-tail of AS hops.

Tier 1 Peerings

  • Peering Matrix - This plot shows a matrix of which networks are peering with each other. Note that all ASNs of Interest are listed along the top, but down the left hand side only the networks are shown for which enough data could be found. This means that for each network on the left, the matrix shows if they have a connection to each network along the top. For the Tier 1 networks, Cogent and Hurricane Electric aren’t peering (a dispute that started in 2009 and is still ongoing!), which means they must be using one of the other Tier 1 networks to reach each other (which would actually be transit not peering, so they aren’t transit free, and they may be be paying for that too, so not settlement free). This mean they really don’t meet the criteria of being a Tier 1. Or if you’re in the camp of “it’s OK to be missing one network” then AS1273 and AS1335 are both Tier 1s too, congrats to them! AT&T, Verizon, Orange, and Liberty Global, some of the incumbent telcos claiming to be Tier 1s, are only connected to 16, 18, 24, and 24 (respectively) of all the networks examined (the most connected networks in the world).

  • v4 Prefixes where Fewest AS Hops Is (Not) via Any Tier 1 - This plot shows the percentage of IPv4 prefixes each network can reach, where the shortest AS path that networks sees doesn’t start with a Tier 1 ASN. This is to show how reliant each network is on their Tier 1 peers and upstreams. The networks which push to be either a global IP carrier, or a really well connected network due to the nature of their business, can reach the majority of prefixes without having to use any of their Tier 1 peers or upstreams. 5 out of 9 networks who have >= 50% of IPv4 prefixes with a shortest path not via a Tier 1 peer, are not Tier 1s, only 4 of them (HE, Telia, Cogent, and Lumen) are Tier 1s. Again, the networks which rely heavily on their Tier 1 peers are the Tier 1s which grew out of being an incumbent telco.

  • v6 Prefixes where Fewest AS Hops Is (Not) via Any Tier 1 - This plot shows the same as above, but for IPv6. Most networks are in roughly the same place +/- a few positions, except for Bharti Airtel, they dropped quite a bit compared to their IPv4 connectivity. AS20473 is missing in the IPv6 plot because a full IPv6 table wasn’t available for them.

Prefix Visibility

  • % of All Announced v4 Address Space Visible via Each Network - This plot shows how much of the IPv4 space each network has reachability to. It’s not enough to compare the number of prefixes each network sees because they see different but overlapping prefixes. Instead the address space seen across all networks is collected, and each network is checked to see what percentage of this total address space they have visibility of. A full IPv4 table was not available for AS57463 so their result is expected/justified.

  • % of All Announced v6 Address Space Visible via Each Network - This plot shows the same as the previous plot, but for IPv6 address space. A full IPv6 table was not available for AS20473 so their result is expected/justified. One would expect most networks to have circa 99% visibility of all announced address space (100% is not achievable due to various reasons such as geopolitical issues, sovereignty, prefix aggregation, broken filters, and so on.), but the visibility of IPv6 space for each network is notably less than for IPv4. In a shocking state of affairs; the networks with the lowest v6 visibility are, well, you know who…

  • % of v4 Prefixes (Not-)Visible by Other Networks - This plot shows what percentage of v4 prefixes each networks has visibility of, which are not visible by any of the other networks which were examined. Clicking “% Prefixes Visible by Other Networks” in the legend will remove that data, making the remaining data easier to see. This should be close to 0% for every network, but will never be exactly 0% for every network (for a range of reasons).

  • % of v6 Prefixes (Not-)Visible by Other Networks - This plot show the same information as the previous plot, but for IPv6 prefixes. Clicking “% Prefixes Visible by Other Networks” in the legend will remove that data, making the remaining data easier to see. It’s interesting to see that for IPv6, as with % of All Announced v6 Address Space Visible via Each Network, reachability of IPv6 is lagging behind IPv4 and by a non-trivial amount.