Avoiding Costs from Oversizing Data Center and Network Room Infrastructure

Overview

The planning model assumes a 10-year design life (Table 1 notes 10-15 years is typical) and an expected load that ramps linearly to its final value by halfway through the lifecycle. The survey finds expected start-up load typically at 30% of design capacity and ultimate expected load at 80-90%, while actual start-up load is typically 20% and ultimate actual load about 60%. That means oversizing of over 1.5x ultimately and about 5x at commissioning.

A typical 100 kW data centre has power and cooling capital cost of about $870,000 ($8.70 per Watt), of which about 40% ($344,000) is wasted by oversizing. Annual maintenance is about 10% of capex; idle loss of the power system is about 5% of rating, rising to 10% including cooling, wasting about 1,500,000 kWh (~$150,000) over 10 years for 100 kW. Total excess lifecycle cost averages approximately 30% of power and cooling infrastructure cost.

Up-front overbuilding is a rational response, because traditional infrastructure is hard to expand and load growth is unpredictable. But if installed capacity were routinely set at the expected ultimate utilisation of about 50%, almost 40% of data centres would be unable to meet their load during their lifetime. The paper therefore recommends adding capacity in step with load.

The remedy is modular, pre-engineered capacity: UPS, batteries, PDUs, rack distribution and row cooling. Main switchgear, main distribution panels and standby generators are typically deployed up front at ultimate capacity, while UPS, batteries, PDUs, bypass switchgear and rack wiring are added in modules. In a modular architecture over 70% of the power system can track load growth.

Key points

1. The model assumes a 10-year design life and load ramping linearly to its final value by mid-life; typical design life is 10-15 years.
2. Survey: expected start-up load is 30% of design and ultimate 80-90%; actual start-up is 20% and ultimate about 60%.
3. Oversizing is over 1.5x ultimately and about 5x at commissioning.
4. A typical 100 kW facility has about $870,000 ($8.70/W) of power and cooling capex, about 40% ($344,000) of it wasted.
5. Maintenance is about 10% of capex per year; idle loss is about 5% of rating (10% with cooling), about 1,500,000 kWh (~$150,000) over 10 years at 100 kW.
6. Total excess lifecycle cost from oversizing averages about 30%.
7. Sizing to the expected 50% ultimate utilisation would leave almost 40% of data centres unable to meet their load.
8. In a modular architecture over 70% of the power system tracks load growth; switchgear, main panels and generators are built up front at ultimate capacity.

Using it in design

• In the capacity planner keep design, installed and expected-load curves separate per phase and show utilisation over a 10-year horizon; default start-up load to 20-30% of design and final load to 60-90%.
• Warn when day-one installed capacity of a modular-capable subsystem (UPS, PDU, CRAH or row cooler) exceeds about 2x the expected first-year load.
• In the phasing template size MV/LV switchgear, main panels and generators for ultimate capacity, and schedule UPS modules, batteries and PDUs as phase additions with pre-provisioned breakers.
• Include idle loss (about 5% of rating, 10% with cooling) and maintenance (about 10% of capex per year) on unused capacity in the TCO calculator.

Who should read it

Designers setting a capacity phasing strategy and managers approving the investment.

Publication details

Authors

Neil Rasmussen

Publisher

Schneider Electric

Year

2012

Document number

WP 37

Document language

English

Source type

Manufacturer

Access

Free to read

Licence

Copyright retained; citation only — the publisher’s terms apply to the original document

Citation