You are here: Home > Publications > RIPE Labs > Mirjam Kühne > An Update on De-bogonising
Content by this author

An Update on De-bogonising

Mirjam Kühne — Feb 21, 2013 01:45 PM
Contributors: Kjell Leknes, Rene Wilhelm
We continued RIPE Atlas measurements to see if we can still find cases where is filtered out.

In The Curious Case of 128.0/16, we looked at how much filtering of is going on on the Internet. The software on some Juniper routers blocks as martian, even though RFC 5735 and RFC 3330 say that this space should no longer be reserved as special address space.

As described in De-bogonising, the RIPE NCC has been helping network operators to clean up these redundant settings.

The last IPv4 addresses we distributed before reaching the last /8 were addresses from During this time, measurements could no longer be made even though we continued to contact those who were filtering this prefix. On request from some of our members who received addresses from this range, we re-started our measurements.

In Q3 of 2012, we started measuring this again by sending pings and traceroutes from two different sets of about 1,000 RIPE Atlas probes to two prefixes in two different Local Internet Registries (LIRs) that offered to help with this project.

Thanks to these measurements, we can now see how (failing) traceroutes to compare to traceroutes to other prefixes in the same LIR's network.


We found that over 95% of the RIPE Atlas probes can now reach destinations addressed with addresses from 64 RIPE Atlas probes failed to get a ping reply from the targeted address.

While bringing the filtering practices down to this level is already an achievement, the ideal situation would be a 0% failure rate. Even if only a few networks filter, it can make operations difficult for those LIRs that received addresses from this range.

Please double check that you are not filtering this prefix and that your router software is up to date. You can test this against or and check if you get a reply.

Measurement results

The graphs below show the result of the measurements done by a set of RIPE Atlas probes to two addresses in assigned to two different LIRs in different parts of the world and the Internet. To eliminate issues specific to a probe that are not related to filtering of, we only consider probes that were able to ping to a control address in the same LIR.

The green points show the percentage of probes succeeding in pinging the address. The red points show the number of probes failing to get a ping reply from the target. In both cases the probes were succesful in reaching the control address.

To eliminate the effects of short-lived fluctuations, success is defined as a probe receiving at least one reply from the target in a one-hour time frame.

De-bogonising Measurements 1Figure 1: Success and failure rate of RIPE Atlas probes to target 1


De-bogonising Measurements 2 Figure 2: Success and failure rate of RIPE Atlas probes to target 1




User Image
Leo Vegoda says:
Feb 21, 2013 04:33 PM
Mirjam, can you also publish graphs showing how these prefixes compare against more established address space? We have seen people accidentally block everything in 192/8 rather than just 192.168/16, so I expect that even long established prefixes don't reach 100% availability.
User Image
Rene Wilhelm says:
Feb 21, 2013 05:37 PM
Leo, you are right that long established prefixes may not have 100% reachability either. With RIPE Atlas we also have seen probes which, for whatever reason, temporary or long term, don't receive any reply from pings to an established destination address. To help eliminate those effects, the two LIRs also provided pingable targets in 185/8 and 217/8 respectively. In the analysis and graphs above we only considered the Atlas probes which could ping those control addresses succesfully. The failures, non reachability, thus really relates to the target addresses being in 128.0/16 space.
Add comment

You can add a comment by filling out the form below. Only plain text is possible. Web and email addresses will be transformed into clickable links. Comments are moderated so they won't appear immediately.