28 July 2026
A new version of the IXP Tracker includes a number of changes, including how we interpret the source data we use from PeeringDB. If you want to see what's changed, the underlying code for the IXP Tracker is open source.
Improvements
More accurate network counts based on registration country
Previously, when detecting if a network is defunct, we weren't using the most current registration country in NRO stats for an autonomous system (AS). This meant that we weren't necessarily removing members from IXPs when they stopped being active.
We updated how we detect this, which means better accuracy, including across historical data. This could reduce the member count for many IXPs, especially for current or recent data.
Updated the way we detect defunct networks, which was affecting AS 112
The AS112 project uses country code ZZ, since it's not located in one particular place. But we have been using ZZ as a way to find defunct networks and remove them. As a result, AS112 has never been listed as a member of any IXPs. We've fixed this across current and historical data.
Bug Fixes
Addressed how we measure non-continuous membership
We have clarified our understanding of how PeeringDB stores info about ASes being members of IXPs. They store this data potentially as multiple records (each record representing one IP v4/6 address pair) and each record holds the date that record was created in PeeringDB. Because records can be "soft deleted" in PeeringDB, these individual IP address entries can appear and disappear in the data visible in PeeringDB.
Until this update, we inferred membership start dates by aggregating all the IP address entries for an AS/IXP combination, and using the date the first record was created as the "start_date". However, this wasn't as accurate as it should be, so now we record membership by using the dates those AS and IXP combinations appear in PeeringDB. This means that both current and historical IXP member counts have changed.
Some members have been added and some removed so the net effect is that some IXP member counts (and hence the aggregated per-country counts) have gone up and some have gone down. Because we determine whether an IXP is active based on the number of members it has, this can also affect IXP counts at country level.
Fixed CSV formatting of IXP info API endpoint
The CSV formatting of the IXP info in the API was returning a mix of CSV and JSON formatted data. We have fixed this so that the list of member ASNs is now correctly formatted.