Summary

  • ARIN’s 2014 report records 901 minutes of unplanned provisioning downtime, including two UPS entries totalling 619 minutes. A spring office UPS failure accelerated the move of critical customer-facing hardware.
  • Mark Kosters’s team tested services, standardized configurations with Puppet and opened the Ashburn site to the public near 9 p.m. on 1 November 2014, after solving a multicast problem on new switches.

The most persuasive argument for moving a registry out of an office park arrived without a slide deck. It arrived as a failed UPS. In ARIN’s 2014 annual report, Chief Technology Officer Mark Kosters says the spring failure sent Operations into a push to relocate critical customer-facing hardware. The same report gives the old arrangement a number: 901 minutes of unplanned provisioning downtime during the year. Two UPS entries account for 499 and 120 minutes; Postgres and a Cox connection account for the rest. That is a service-specific tally, but it made the physical problem hard to romanticize.

ARIN’s headquarters sat in Chantilly, near Dulles Airport. Kosters’s first-person account describes a single power feed into the office park, only two network providers and no sustainable power after a failure. Road construction nearby had already produced unplanned outages. You can write “availability” in a strategy document; you cannot add a second power feed with a better adjective.

The engineering response was a move to an Ashburn colocation facility. Kosters says the team tested services before the transfer, standardized environments and used Puppet for configuration management. Roughly a year of preparation ended on 1 November 2014. The new Juniper switches supplied one moving-day joke nobody had ordered: odd multicast behaviour. The team resolved it and opened the public front door close to 9 p.m. ARIN’s annual report confirms the date and says Ashburn had proved a reliable home for core operations by the time it was written.

This was a concrete choice about where the registry’s working systems could survive a bad day. The address records matter only while the machinery for updating and serving them remains reachable. Kosters moved that machinery toward stronger power and connectivity, while keeping the cutover itself a testable engineering event.