HN-RP-014 · · Engineering · 13 min read · 2,655 words · Version 1.0

Three Minutes, and Where to Keep Them

Distributed battery backup in an 800 volt direct current hall

What ride-through is actually for. Why the battery moved twice before it settled. What lithium iron phosphate gives up, and what it buys back.

Abstract

A data centre battery is not there to run the building. It is there to hold the load steady for the few seconds between losing the utility supply and a generator accepting it. This paper sets out how much time that actually takes, why HyperNext provides about ten times it rather than the hour a conventional design would, why the battery ended up inside the rack power panel after two changes of mind, and why the cell chemistry was chosen on thermal behaviour rather than on energy density.

Contents

  1. 011. What the battery is actually for
  2. 022. Where it sits, and why it moved twice
  3. 033. The chemistry argument
  4. 044. The drawer, the module and the hot swap
  5. 055. Following Open Compute at the connector
  6. 066. What three minutes costs, and what thirty would
  7. 077. Keeping the cells at the right temperature
  8. 088. References and standards

Request the full paper

The complete paper, including all figures, tables, references, and citations, is available as PDF. Enter your details to receive it.

Request paper · HN-RP-014.pdf

Key findings

  • The worst case the battery has to cover is detection, generator start and load acceptance, which is under twenty seconds. Three minutes of stored energy is a margin of roughly ten, not a reserve to run on.
  • Energy, floor area, mass and cost scale linearly with autonomy. Every additional minute is bought at the same price as the first, and buys less.
  • Lithium iron phosphate has lower energy density than the alternatives and a substantially higher thermal runaway onset. For equipment that sits in an occupied hall for fifteen years, the second property decides it.
  • The battery moved from the panel to the rack and back again. The deciding argument was that one drawer design can back both direct current racks and the alternating current racks fed through in-rack conversion, which in-rack batteries cannot do.
  • Hot-swappable modules inside the drawer mean a cell failure is a module exchange rather than a loss of the drawer, which changes what redundancy has to be designed for.