Data Center Projects: Growth Model

Overview

The IT load profile is stated with four parameters rather than one prediction: maximum final load, minimum final load, initial load and ramp-up time. Step size and margin describe how the infrastructure is deployed, and together they form the system capacity plan. Data centre lifetime is typically set at ten years for the growth projection. The paper also gives conversions, kW = BTU/hr x .000293 and kW = tons x 3.52, and notes that each watt of IT power becomes one watt of heat, so the IT cooling profile equals the IT load profile.

Non-scalable elements such as room size, service entrance, utility switchgear and heat rejection (for example chillers) are sized up front for the maximum final load. Scalable elements, racks and rack-based power and cooling, are phased in from the initial load. The paper observes that the typical data centre runs at less than half capacity and that many finish their lifetime at less than half the assumed maximum load.

The phasing strategy follows from uncertainty. If minimum equals maximum with a short ramp-up, full upfront build-out is preferred; great uncertainty justifies a low initial build-out with more, smaller phase-in steps. Step size is set by the scalability of the system architecture, the room layout (typically row-based steps), uncertainty about IT load and facility lifetime, and the cost and disruption of each step.

For total cost of ownership comparisons, the 'expected' final load defaults to the average of maximum and minimum final load, while the capacity plan must still be able to cover the maximum. This corrects analyses that assume full build-out and so hide the waste of oversizing. The model is qualitative: it offers a planning rule, not a data-centre statistic.

Key points

1. Six parameters: maximum final load, minimum final load, initial load and ramp-up time (load profile); step size and margin (capacity plan).
2. Data centre lifetime is typically ten years for the growth projection.
3. kW = BTU/hr x .000293 and kW = tons x 3.52; each IT watt becomes a watt of heat, so the cooling profile equals the load profile.
4. Non-scalable elements (room, service entrance, switchgear, heat rejection such as chillers) are sized up front for the maximum; racks and rack-based power and cooling phase in from the initial load.
5. The typical data centre runs at less than half capacity; many end their life below half the assumed maximum load.
6. Minimum equal to maximum with short ramp-up favours full upfront build-out; great uncertainty favours a low start with more, smaller steps.
7. For TCO the expected final load defaults to the average of maximum and minimum; the capacity plan must still cover the maximum.
8. Step size depends on architecture scalability, room layout (typically row-based), load and lifetime uncertainty, and cost and disruption per step.

Using it in design

• Capture the IT load profile at project setup as four fields (initial kW, minimum final kW, maximum final kW, ramp-up years, default lifetime 10 years), not one design load.
• Tag every catalog element as scalable or non-scalable; size non-scalable ones (utility service, switchgear, chillers, room) to maximum final load, plan scalable modules in steps with a stated margin, and plot capacity against load over time.
• Warn when the min/max spread is large but the design has a single upfront build.
• In the TCO module default expected load to (min + max)/2 and show the oversizing waste of upfront build-out; keep cooling in kW with BTU/hr and ton conversions.

Who should read it

Planners preparing phased capacity plans and TCO analyses, and owners making the investment decision.

Publication details

Authors

Neil Rasmussen; Suzanne Niles

Publisher

Schneider Electric

Year

2012

Document number

WP 143

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