Summary

  • Remi Hendriks told RIPE 92 that his team's anycast test bed had points of presence in Lagos and Johannesburg. Some neighbouring African networks still reached its Paris node rather than Lagos.
  • In that deployment, the median round-trip time from African networks was almost twice the global median. Hendriks calls the result preliminary and asks other operators to compare their catchments; local peering, transit choices and operating constraints still determine which node a route reaches.

An anycast operator can buy a server in Lagos. It cannot buy the routing decision of every network around it. Remi Hendriks of the University of Twente described that collision at RIPE 92: his team's test bed placed points of presence in Lagos and Johannesburg, while some networks near the Nigerian site continued to send their traffic to the team's Paris point.

Anycast announces one service address from several locations. BGP selects a path to one of them. The commercial case for an African node assumes that nearby users will reach it, reducing delay and leaving another site available during trouble. Hendriks looked at the actual catchments—the networks served by each point—and found a geography that did not follow the map. Networks from several former French colonies appeared in the Paris catchment despite the Lagos node. His team measured a median round-trip time from African networks to this test bed almost twice the global median.

That result belongs to this deployment. Hendriks offered limited cross-border peering as a likely explanation, and the official talk abstract frames it as work in progress. Operators in the question period supplied further mechanisms. A root-server operator described political controls, monopolies and the difficulty of maintaining local relationships. Another participant pointed to the economics of existing international transit: networks may keep using a cable and provider path toward Europe even when a neighbouring-country path has lower delay.

The sharpest point is operational. A nearby node delivers little to a network whose route to it is unattractive or unavailable. More servers can multiply the number of pins on an operator's map without changing the catchment. Hendriks asked other anycast operators to share observations and collaborate on deployment. Dated path and catchment measurements before and after local peering changes would show which networks move from Paris to Lagos and how much delay they save.

The placement decision is only half the investment. The other half sits with exchange points, transit contracts, cross-border links and the people who can operate them. Hendriks has put a test node on the ground; the route to that node is still a negotiation.

Sources: RIPE 92 primary transcript · official talk page · official RACI profile · University of Twente's meeting account.