Robert Kisteleki Put a RIPE Atlas Anchor in a VM. The Rack Survived.

Summary

• Kisteleki's 2017 plan proposed RIPE Atlas anchors in public-cloud and partner-hosted virtual machines, reaching networks awkward for physical hardware. A May 2018 update named the pilot hosts and the installation work.
• The July outcome found no discernible performance difference in the data analysed from a nearby VM and physical anchor. Stephen Strowes's RIPE Labs comparison examined seven days of measurements; the RIPE NCC's 2019 report later recorded virtual anchors extending the network's reach.

The sales pitch for a global measurement network is wonderfully simple: put a vantage point where you need to see. The invoice arrives in cardboard. A physical anchor needs a host willing to install and maintain a box. Kisteleki's 2017 proposal asked whether some of those vantage points could arrive as virtual machines instead, especially where placing a device was difficult.

He designed two tracks. One would try public cloud providers; the other would work with willing partners who could supply a VM in their own networks. The virtual anchors would join the ordinary RIPE Atlas anchor mesh, be measured by other anchors and act as probes themselves. A tag would let researchers include or exclude them deliberately. That tag matters: a cloud vantage point and a box in a network are both useful, but a researcher should know which one answered.

By May 2018, the experiment had an RIPE NCC VM and installations with Sander Steffann, Jared Mauch and Amazon. The useful comparison came from DigitalOcean, which hosted a VM near a physical anchor. In Kisteleki's July report, the analysed performance showed no discernible difference. The report credits Iñigo Ortiz de Urbina Cazenave, Michela Galante and Alun Davies as contributors; this was team work with volunteer hosts.

Strowes put numbers behind the comfort. His RIPE Atlas analysis used seven days of mesh data involving roughly 300 other anchors. HTTP response medians sat near zero difference; small ping differences appeared, while address family and network effects often mattered more. The result gave the VM a practical audition, with a defined pair and a defined week.

Then came the questions a benchmark cannot answer. Should the host hand RIPE NCC a ready VM, or should the registry build one itself in a strategically useful location? May a host place an anchor in a network it does not own? How many VMs can the system accept before storage, processing and credits become a problem? Will a host who never bought a box keep caring when the VM goes quiet? Kisteleki's outcome laid those choices on the table and proposed beginning with the lighter host-provided option.

The RIPE NCC 2019 annual report says virtual anchors had extended Atlas's reach. The pilot had crossed into operation. Kisteleki's important move was to make the virtual machine earn its place twice: once in the measurements, then in the rules for who runs it. The rack survived; the responsibility still needed a name.