Analysing impact of NTT & Cogent de-peering


Over the past few days, there’s been an active discussion in the networking community around Cogent and NTT de-peering. Cogent & NTT have turned down their BGP sessions in Europe and this is forcing European local traffic between NTT & Cogent customers/downstream via the US.

Now, peering is an extremely commercial issue and I am going to refrain from doing any commentary about anything but the technical aspect of it. Focusing purely on the technical aspect of it, I looked at the routing table to see if I could find out the answer to a fundamental question: Which of the downstream ASNs of each network would be impacted if a complete de-peering happens?

Impact of complete de-peering

It’s a known case that NTT and Cogent were peering in Europe & US (only). Now with de-peering already happening in Europe, they are only left with BGP sessions in the US. A quick look at the routing table & BGP communities reveals that presently NTT and Cogent are peered in the following locations:

  1. Chicago
  2. Dallas/Houston
  3. Los Angeles
  4. Seattle
  5. Palo Alto/San Jose
  6. Boston

If this dispute escalates further and a complete de-peering happens in that case both networks will end up having a blackhole. Customers sitting on either side (and their single-homed downstreams) will not have any routes to each other. When transit-free networks de-peer there is no alternate route and that results in blackholing of traffic. Downstream ASNs as long as are also customers of any other non-NTT/non-Cogent network will be able to reach either side. But if someone is multi-homed with routes ultimately converging on either of these networks would have an issue.

To determine who will be impacted, one has to look at single-homed customers of both networks and then analyse all the ASNs in the path.

How to find single-homed customer of NTT (AS2914) and Cogent (AS174)?

This process can be somewhat challenging because both NTT and Cogent provide their full routing tables to various route collectors. A full table includes both their downstream customer routes and their peer routes. So one has to exclude peer routes to get a list of downstream. Peer routes are not something that is directly or easily guessed. A starting point here can be Wikipedia tier 1 network list and assume networks in the list - AT&T, Verizon, Tata Comm, GTT etc are not downstream of either NTT or Cogent. I might still identify some ‘peers’ who are not listed (who I know are peers!) But that’s still okay because by definition if a network is a peer of NTT or Cogent, it must have other direct/indirect routes to the rest of the routing table. So those anyway would get filtered out in my lookup. Additionally, I expanded the downstream AS cone for each network to include both direct and indirect downstreams. This gives me a list of 52181 ASNs (single + multi-homed) which use Cogent AS174 and 28865 ASNs (single + multi-homed) which use NTT.

Now, to determine which of these are single-homed, let’s examine routing tables from various transit-free networks, as listed on Wikipedia

ASN AS Name Routing dump
3320 Deutsche Telekom Global Carrier RIPE RIC RRC01
6830 Liberty Global RIPE RIC RRC01
5511 Orange / France telecom RIPE RIC RRC01 (downstream table only)
3491 PCCW Global RIPE RIC RRC01
1239 Cogent / Sprint RIPE RIC RRC01
6461 Zayo RIPE RIC RRC01
12956 Telefónica Telxius RIPE RIC RRC03
6453 Tata Comm RIPE RIC RRC03
6762 Telecom Italia Sparkle / Seabone RIPE RIC RRC12
3356 Lumen
701 Verizon Not found in RIPE RIS or Oregon route-views

The idea here is that if any downstream ASNs of NTT or Cogent are visible in the AS_PATH table for the listed ASNs, then those ASNs are not single-homed and are excluded. This provides me with a subset of ASNs that are currently entirely dependent on either NTT or Cogent, either directly or indirectly, to reach the rest of the default-free zone


Network Name ASN Number Single-homed downstream ASNs List of Single-homed downstream ASNs
NTT AS2914 211 Click here
Cogent AS174 1084 Click here

Limitations of the data:

  • ASNs here will broadly have a lack of connectivity to ASNs on the other side. There can still be some private PNIs, peering over IXPs resulting in connectivity though mostly it is localised and chances of a network in say Japan in single-homed downstream of NTT peering with a network in EU which is single-homed behind Cogent is quite low.
  • There is a lack of visibility from Verizon AS701 due to no public dump as well as Orangle AS5511 since only the downstream table is shared. Thus a route with path 5511 701 X or 701 5511 X could exist and would have an alternate route to the other side. Those cases are simply being missed in this lookup.
  • Routing table doesn’t always mean capacity. A network might be multi-homed with extremely low capacity from an alternate path. It won’t show up in my data but would be impacted.
  • This data is based on a specific lookup I did at 06:00 IST (GMT+5.5) on 20th Feb. Routing table changes rather quickly. Massive changes are unlikely but one may find ASNs in the list that later become multi-homed or vice versa.

Does it matter?

Well yes and no. Transit-free networks are essential for ensuring internet connectivity, allowing packets to be transmitted to any network globally. While they offer broad reach, their share of total traffic volume is relatively limited. Nowadays, the bulk of internet traffic originates from leading content providers and content delivery networks (CDNs), primarily directed towards end-user networks rather than between transit-free networks.

In terms of traffic volume, the flow on these paths for eyeball networks on either side would be minimal. However, content networks utilizing NTT and Cogent as transit providers would likely experience the impact.

Disclaimer: The views expressed in this blog post are solely mine and do not necessarily reflect the views of my employer.