You are here: Home > Publications > RIPE Labs > Robert Kisteleki > The AMS-IX Outage as Seen with RIPE Atlas

The AMS-IX Outage as Seen with RIPE Atlas

Robert Kisteleki — 14 May 2015
Contributors: Vesna Manojlovic, Emile Aben, Daniel Karrenberg
During the RIPE 70 Meeting in Amsterdam this week (on 13 May around 10:00 UTC), we experienced a network outage at AMS-IX. Let's see how this was monitored by various tools.

 

The outage was nicely visible on the AMS-IX traffic staticstics (see Figure 1 below).

 

AMS-IX Statistics Figure1: Daily IPv4 traffic graph showing the outage (source: AMS-IX)

 

This was a good opportunity to monitor this with various tools the RIPE NCC makes available.

The RIPE Atlas seismograph tool visulises multiple ping measurements. In the three figures below you can see ping measurements towards three different RIPE Atlas anchors, all located in Amsterdam: one in the RIPE NCC network, one at Surfnet and one at Afilias.

seismograph Figure 3: RIPE Atlas seismograph (RIPE NCC)

Seismograph Surfnet Figure 4: RIPE Atlas seismograph (Surfnet)

Seismograph Afilias Figure 5: RIPE Atlas seismograph (Afilias)

You can clearly see the time during which very few ping measurements were successful from the RIPE Atlas probes. Some of the probes could still get to it, because there was some kind of network connectivity. They probably did not go through AMS‑IX, but this is mostly speculation. This particular tool can clearely show that there was a problem and that the problem was closer to the RIPE Atlas anchor rather than the probes.

Figure 6 below show the same event in DNSMON measuring towards K-root.

DNSMON Figure 6: DNSMON showing ...

On the Y axis you can see a list of RIPE Atlas anchors from all over the world. During the interruption, even though they were still collecting data, they couldn't connect to our infrastructure, so they couldn't send data to the RIPE Atlas storage . That causes the white "no data" gaps in the image. Eventually, when the connection was restored, they started to send all the delayed reports -- but not neccesarily in order, so the graph contain spots. When the data transfer is finished, the gaps are expected to disappear. This means that by now these gaps are gone .

Of course it would be very nice if we could use RIPE Atlas data to understand such events while they are happening. We are doing research on algorithms for real time analysis which would allow this in the future based on the RIPE Atlas streaming interface .

0 Comments

Add comment

You can add a comment by filling out the form below. Comments are moderated so they won't appear immediately. If you have a RIPE NCC Access account, we would like you to log in.