Failover And Transfer

When the burner trips or is serviced, transfer switches must hand the load to the battery and grid fast enough that the compute never sees a gap.

Handing off without a gap

Failover is the act of switching the load from a failed source to a healthy one. In a data center this is done by automatic transfer switches coordinated with the UPS. The requirement is strict: the transfer must complete, or the UPS must bridge it, within the tolerance of the load, which for accelerators is very short.

This sequence is the operational heart of answering the availability gate. Because the burner will trip and will be serviced, the value of the hybrid depends entirely on the transfer being invisible to the compute. The UPS/battery covers the instant of detection and switching; the transfer then selects the grid or a spare burner unit; the load rides through without a checkpoint loss.

Transfer equipment is itself a possible point of failure, and the availability math warns that shared switchgear can create common-mode failures that defeat redundancy. So the design uses well-characterized transfer schemes, avoids single shared choke points, and tests failover regularly — the same discipline mature data centers apply.

Done right, failover converts a burner trip from an outage into a non-event, which is precisely how a 0.86-0.995 base source becomes part of a Tier III campus.

• Failover switches load from failed to healthy source
• UPS bridges the switching instant
• Transfer selects grid or spare burner unit
• Transfer hardware must avoid common-mode failure