Improving RIPE Atlas Coverage - Which Networks Are Missing?

RIPE Atlas has already deployed to an impressive number of networks. A total of 2,997 (5.9%) IPv4 ASNs and 1,110 (11.4%) IPv6 ASNs are covered by RIPE Atlas. But are RIPE Atlas probes deployed in the networks with the largest user bases?

Recently APNIC Labs started publishing estimates of user populations. Though this is preliminary work, it is the best publicly available dataset I am aware of that shows what part of the user market a particular ASN covers in a country. The data has a few known issues [1], but as a first estimate of what portion of a country's user base is behind which ASN, this is highly useful.

RIPE Atlas coverage in eyeball networks

When combining the APNIC Labs user population estimates and RIPE Atlas deployment numbers, we can try to assess which large eyeball networks are not yet covered  by the RIPE Atlas infrastructure. In Table 1 we show the number of RIPE Atlas probes for the top 20 networks (by estimated user population). What the table shows is large differences in RIPE Atlas' deployed base. The full results of combining this data can be found here . One general observation from the full list is that RIPE Atlas generally covers networks in Europe and North America well, but for some of the largest networks, by the estimated user population, we have either no online probes, or very low coverage.

Table 1: Top 20 eyeball networks by estimated number of users and their coverage with RIPE Atlas

This information shows us where to direct our efforts to deploy more RIPE Atlas probes. For networks with no probes online, we definitely want RIPE Atlas probes deployed. For instance, large access networks in China, India and Nigeria are not well covered at the moment. This is useful information for RIPE Atlas ambassadors who help to increase the reach of the RIPE Atlas infrastructure.

Other applications of this data: eyeballtrace

There are other interesting applications of these user population estimates. For instance, how about automatically selecting a few RIPE Atlas probes to do traceroutes from networks with a significant user population in a given country? This can be useful if, for instance, you are trying to decide where to host content, or if you are trying to optimise routing for content specifically aimed at users in a certain country. Another use case could be to quickly check whether paths between your network and users in a specific country are affected by a route leak, hijacking or other Internet routing event.

As a proof of concept I've created an "eyeballtrace" command-line tool. As input, it takes a country and an Internet host (either by hostname or IPv4 address). The output are traceroutes from RIPE Atlas towards the intended destination from all ASNs that have an eyeball market share of over 1% in that country. The output is sorted by market share, and currently a single traceroute per ASN is displayed. This provides a rough sketch of routing from eyeball networks in a given country.

Let's look at a (very) hypothetically example. Say I'm in Armenia and I'm considering hosting content in Singapore. How will users be routed to that content? Or I'm a content hosting provider in Singapore, and I want to start targeting the market in Armenia. How will users be routed to my hosting location?

The output of eyeballtrace -c AM sg-sin-as133165.anchors.atlas.ripe.net is listed below. 

 $
 
  eyeballtrace -c AM sg-sin-as133165.anchors.atlas.ripe.net
 
 

 

 Please wait, while RIPE Atlas performs traceroutes to: sg-sin-as133165.anchors.atlas.ripe.net
 

 

 

 ## AS12297/ARMENTEL - ArmenTel CJSC (40.4% of market in AM)
 

 NO RIPE Atlas coverage!!
 

 

 ## AS44395/ORG-UL31-RIPE - UCOM LLC (17.6% of market in AM)
 

 #prb:10609 dst:sg-sin-as133165.anchors.atlas.ripe.net
 

 1 (AS44395) host-137.192.162.46.ucom.am [0.371, 0.38, 0.451] ||
 

 2 (AS44395) host-105.192.162.46.ucom.am [0.446, 0.549, 1.107] ||
 

 3 (AS8932) host-21.240.48.185.ucom.am [0.799, 2.118, 2.421] ||
 

 4 (AS8932) host-34.240.48.185.ucom.am [29.506, 29.521, 29.535] ||
 

 5 (AS1299) sfia-b2-link.telia.net [29.491, 29.512, 29.514] |Sofia,Sofia-Capital,BG|
 

 6 (AS1299) win-bb2-link.telia.net [50.101, 50.112, 51.649] |Vienna,Vienna,AT|
 

 7 (AS1299) ffm-bb2-link.telia.net [62.564, 62.587, 66.44] |Frankfurt am Main,Hesse,DE|
 

 8 (AS1299) ffm-b12-link.telia.net [69.1, 70.614, 70.63] |Frankfurt am Main,Hesse,DE|
 

 9 (AS1299) ntt-ic-155239-ffm-b12.c.telia.net [62.927, 62.959, 63.064] |Frankfurt am Main,Hesse,DE|
 

 10 (AS2914) ae-5.r21.frnkge03.de.bb.gin.ntt.net [69.781, 73.022, 75.485] |Frankfurt am Main,Hesse,DE|
 

 11 (AS2914) ae-3.r22.londen03.uk.bb.gin.ntt.net [77.292, 77.326, 77.348] |London,England,GB|
 

 12 (AS2914) ae-6.r21.sngpsi05.sg.bb.gin.ntt.net [329.452, 329.553, 359.494] |Singapore,,SG|
 

 13 (AS2914) ae-13.r00.sngpsi02.sg.bb.gin.ntt.net [325.673, 325.989, 328.365] |Singapore,,SG|
 

 14 (AS2914) 116.51.27.190 [339.515, 339.595, 341.498] ||
 

 15 (AS133165) 103.253.144.230 [326.872, 326.902, 327.081] ||
 

 16 (AS133165) 103.253.144.218 [339.191, 339.239, 339.343] |Singapore,,SG|
 

 

 

 

 ## AS49800/GNC-ALFA - GNC-Alfa CJSC (14.1% of market in AM)
 

 #prb:12503 dst:sg-sin-as133165.anchors.atlas.ripe.net
 

 1 () 192.168.1.1 [0.603, 0.611, 0.619] ||
 

 2 (AS49800) 46.19.101.129 [0.963, 1.092, 1.176] ||
 

 3 (AS49800) ip-46-19-97-65.gnc.net [3.091, 3.146, 3.195] ||
 

 4 (AS49800) lt-5-3-0-1003.gnc.net [3.771, 3.811, 10.469] ||
 

 5 (AS12389) 92.50.196.101 [34.265, 34.279, 34.821] ||
 

 6 (AS12389) 95.167.95.85 [88.33, 88.511, 88.556] ||
 

 7 err:{u'x': u'*'}
 

 8 (AS4637) i-0-4-0-1.ulco-core02.bi.telstraglobal.net [153.837, 154.276, 155.939] |London,England,GB|
 

 9 (AS4637) 202.84.249.166 [994.137, 1039.179, 1065.954] ||
 

 10 (AS4637) i-0-0-0-0.istt02.bi.telstraglobal.net [768.912, 835.048, 856.667] ||
 

 11 (AS4637) unknown.telstraglobal.net [395.453, 395.942, 400.758] ||
 

 12 (AS133165) 103.253.144.234 [386.701, 390.09, 398.518] ||
 

 13 (AS133165) 103.253.144.218 [411.51, 430.383, 463.378] |Singapore,,SG|
 

 

 

 

 ## AS49363/ORANGE-ARMENIA - Orange Armenia CJSC (12.6% of market in AM)
 

 NO RIPE Atlas coverage!!
 

 

 ## AS43733/K-TELECOM - K-Telecom CJSC (5.6% of market in AM)
 

 #prb:10659 dst:sg-sin-as133165.anchors.atlas.ripe.net
 

 1 (AS43733) 9.3.76.217.in-addr.mts.am [0.346, 0.346, 0.405] ||
 

 2 (AS8932) host-9.241.48.185.ucom.am [0.457, 0.461, 0.469] ||
 

 3 (AS8932) host-34.240.48.185.ucom.am [29.786, 29.789, 29.82] ||
 

 4 (AS1299) sfia-b2-link.telia.net [29.442, 29.474, 29.489] |Sofia,Sofia-Capital,BG|
 

 5 (AS1299) win-bb2-link.telia.net [62.278, 62.29, 62.292] |Vienna,Vienna,AT|
 

 6 (AS1299) ffm-bb2-link.telia.net [60.084, 60.087, 60.096] |Frankfurt am Main,Hesse,DE|
 

 7 (AS1299) ffm-b12-link.telia.net [60.416, 60.442, 61.045] |Frankfurt am Main,Hesse,DE|
 

 8 (AS1299) ntt-ic-155239-ffm-b12.c.telia.net [62.936, 62.969, 63.239] |Frankfurt am Main,Hesse,DE|
 

 9 (AS2914) ae-5.r21.frnkge03.de.bb.gin.ntt.net [60.642, 60.685, 61.497] |Frankfurt am Main,Hesse,DE|
 

 10 (AS2914) ae-3.r22.londen03.uk.bb.gin.ntt.net [74.735, 75.501, 76.379] |London,England,GB|
 

 11 (AS2914) ae-6.r21.sngpsi05.sg.bb.gin.ntt.net [325.654, 325.654, 325.668] |Singapore,,SG|
 

 12 (AS2914) ae-13.r00.sngpsi02.sg.bb.gin.ntt.net [328.489, 328.618, 329.237] |Singapore,,SG|
 

 13 (AS2914) 116.51.27.190 [333.071, 333.072, 333.259] ||
 

 14 (AS133165) 103.253.144.242 [904.237, 1055.309] |Singapore,,SG|
 

 15 (AS133165) 103.253.144.218 [768.114, 850.457, 927.405] |Singapore,,SG|
 

 

 

 

 ## AS47975/KT-AS - Pimox LLC (2.5% of market in AM)
 

 NO RIPE Atlas coverage!!
 

 

 

 openipmap_url: https://marmot.ripe.net/openipmap/tracemap?msm_ids=2066032&show_suggestions=0

This output provides some interesting insights.

Some ASNs in Armenia (AS12297 (ArmenTel), AS49363 (Orange) and AS47975 (Pimox)) didn't have a RIPE Atlas probe online when performing this measurement. This is something we could change by distributing RIPE Atlas probes specifically for use in these networks at our next regional meeting in Yerevan, Armenia .

From the traceroutes you can see that these routes generally go through Frankfurt and London. Latencies to the destination network are roughly in the 300-400 ms range. For AS43733, which has roughly 5% market share in Armenia, we see latencies of over 700 ms. Looking at the traceroute for this last case, a latency jump can be seen in the traceroute output when entering the destination network, which is probably caused by the return path not going through AS2914 but taking a longer route.

At the end of the output there is a URL for an OpenIPMap visualisation of the places we think the paths go geographically. Figure 1 is a screenshot of this visualisation. The OpenIPMap interface also doubles as a way to correct geolocation information on routers in the traceroutes.

Figure 1: OpenIPMap visualisation of an eyeballtrace from Armenia to sg-sin-as133165.anchors.atlas.ripe.net

This proof-of-concept "eyeballtrace" tool is available on GitHub . But please beware that this is a prototype and your mileage may vary.

Conclusion

RIPE Atlas coverage of eyeball networks in North America and Europe is generally good, while we can still improve significantly in the rest of the world. We specifically invite RIPE Atlas ambassadors and other people interested in distributing and hosting RIPE Atlas probes to look at this data and focus the efforts on the networks with low or no coverage as can be seen here .

For the adventurous, we have a proof of concept eyeballtrace tool to look into, which uses APNIC Labs' population by ASN data to do traceroutes from networks with significant user populations.

Footnote s

[1] Known issues on AS population dataset

• ASNs that have significant user populations in multiple countries are currently not accounted for. An example is AS6830.
• Organisations that operate multiple ASNs are not aggregated
• This data is an order of magnitude estimation, and useful in that context. The figures are based on random distribution of APNIC Labs 1x1 tests from users in each economy combined with ITU population and subscriber statistics. The relative percentages are indicative of market share, but the user population counts are more speculative.

[2] Acknowledgements

The idea to combine APNIC Labs AS population data with RIPE Atlas originally came up in an email conversation with Rudolf van der Berg (then OECD, now Tele2). The APNIC Labs AS population data itself is produced by Geoff Huston and George Michaelson at APNIC.