Brownfield & renovation

Taking an existing data center to Tier III — what actually blocks it

A brownfield upgrade rarely fails on the thing the client is worried about. It fails on spare capacity in the plant nobody planned to touch, and on cooling that was never tested against the maintenance requirement in the first place.

What Tier III actually asks of an existing facility

Tier III is Concurrently Maintainable. The requirement is narrow and absolute: each and every capacity component and every element of the distribution path can be removed from service on a planned basis without impacting the critical load. Not most of them. Not the ones you expect to service.

On a greenfield project that shapes the design from the first sketch. On a live facility it is an audit of decisions somebody else made, often a decade ago, usually without this requirement in mind. The building already works — that is exactly what makes the gap easy to miss.

Failure one: the plant you are reusing has less room than you think

The commercial case for a brownfield scheme almost always rests on reuse. The transformers are there. The generators are there. The chilled water plant is there. The scheme is a tap-off, and the tap-off is what makes the numbers work.

So the question that decides the project is a capacity question:

A site-level check cannot answer that. Adding up total installed capacity against total IT load tells you whether the site is big enough; it says nothing about whether this transformer, carrying that legacy load, has room for the new feeders. The only way to answer it is to walk the distribution downward from each reused item and add up the load that actually reaches it.

Worked through, on a 2500 kVA unit at 0.8 power factor:

It fits. It is also finished: 180 kW of headroom is not a second phase, and at 91% the unit has no margin for the load growth every data center sees. The honest answer to the client is not “yes” — it is yes, and this transformer is now a constraint on everything that comes after.

Where the number goes wrong

• Counting the legacy load twice. If the old building is drawn into the model and its demand is also entered as an existing load figure, the plant looks full when it is not. Pick one.
• Using nameplate instead of demand. The legacy building’s connected load is not what it draws. Use measured demand where you have it, and say so.
• Forgetting the mechanical load. Chillers, pumps, CRAH fans and the house load all sit on the same plant. A capacity check that only counts IT is optimistic by the entire mechanical bill.

Failure two: cooling was never in the sweep

Concurrent maintainability applies to cooling exactly as it applies to power. A hall that keeps its power but loses its cooling is still down — only more slowly, and with equipment damage on the way. Yet cooling is the part most often assessed by counting units rather than by testing the removal.

Here is a real facility, anonymised. Three independent power trains, eleven chilled-water CRAH units at 222 kW each, an IT load of 1982 kW. On paper the cooling is N+2: nine units carry the load, eleven are installed.

N+2 on the schedule, and the facility still cannot take a board out of service. The redundancy is real; it is just distributed so that any single board carries more cooling than the spare capacity covers. This is a Tier II facility that everyone involved believed was Tier III.

What fixes it

Two routes, and the cheaper one is usually not the obvious one.

• More units. To survive losing the worst board, total capacity must exceed the IT load plus that board’s share. Balanced over three boards that means fourteen units, not eleven. Three more machines, three more sets of pipework and power.
• Dual-feed the units you already have. Give each CRAH a second source from a different board — in practice a mechanical distribution board with two sources, not eleven separate transfer switches. Lose any board and all eleven units stay running, at full 2442 kW. The cost is cable and a board, not plant.

In the facility above the second route moved it from Tier II to Tier III with no new equipment at all.

The failure mode nobody looks for

There is a third risk specific to renovation, and it is the one that damages reputations: the works quietly break something that used to pass.

Re-terminating a feeder to serve the new hall, moving a load from one board to another, taking a redundant unit out to make room — each is defensible on its own, and each can remove a maintenance path that the existing facility depended on. The design report says “Tier II to Tier III”. It should also say which specific requirements changed state, in both directions.

That means grading the captured existing state and the proposed state against the same criteria, check by check, and reading the regressions as carefully as the improvements.

Phasing: the Tier you have on the way, not only at the end

A live facility is upgraded in stages, and the stages are where the exposure is. A scheme that lands at Tier III on the final day may sit below its starting resilience for months in the middle — while the client is still running production on it.

The question to answer per phase is the same one as at the end: after this stage, what can be taken out of service, and does the reused plant still cope with what is connected by then? Assign every component to a phase, replay the build one phase at a time, and grade each intermediate state.

A checklist for the next brownfield assessment

1. Capture the existing facility as a model before drawing any of the works.
2. For every item being reused, establish what it already carries — measured, not nameplate.
3. Walk the distribution downward and compute spare capacity per device, not per site.
4. Take out each board, each transfer switch and each distribution path in turn. Re-add the surviving power and cooling against the IT load each time.
5. Check the generators are Prime or Continuous rated. Standby-rated units are limited to roughly 200 hours a year and are not acceptable for Tier III without manufacturer attestation.
6. Grade the before and after against the same criteria and list every check that changed state, including the ones that got worse.
7. Repeat the whole assessment at the end of each construction phase.