Specification of Modular Data Center Architecture

Overview

The paper names three elements of modularity: device modularity, subsystem modularity and module linkage, the rules that tie deployments of different subsystems together. Linkage is called the least developed and most valuable. Example rules: 1 rack PDU per rack, 1 PDU per 20 racks, 1 CRAH per 40 racks, 1 data acquisition system per 200 racks, 1 generator per 500 racks. The hierarchy has five levels: facility, room, pod, cabinet, device.

The authors argue that facility size, growth certainty, redundancy goals and site constraints mean a handful of standard architectures, using cabinet, pod or room as the deployment unit, can cover most needs. Typical scales: branch room 1 room/1 pod/5 cabinets/20 devices; small DC 1/4/40/250; large DC 6/30/360/2000. A modularity chart shows which subsystems scale at which level and which modules add capacity versus redundancy.

In the worked example, pod-level subsystems are aligned at 60 kW increments, and 3 pods plus humidifier, chilled-water distribution, pump and lighting form a room. CRAH redundancy is one N+1 module per three capacity modules at pod level, and one humidifier per 12 CRAH modules. In the N+1 chiller plant each chiller supports 4.5 pods: two on day one, a third when the fifth pod is added.

Paralleling devices onto large shared buses forces design of every configuration up front; independent per-pod buses scale best but need devices with internal N+1 redundancy for cost-effective fault tolerance. The paper also lists attributes an architecture should publish: PUE versus load per deployment step, redundancy by subsystem, U per pod, floor loading, cost per kW, Tier, water per kW, and average and peak kW per cabinet.

Key points

1. Three elements: device modularity, subsystem modularity, module linkage; linkage is least developed and most valuable.
2. Five-level hierarchy: facility, room, pod, cabinet, device.
3. Example linkage: 1 PDU per 20 racks, 1 CRAH per 40 racks, 1 generator per 500 racks.
4. Small DC typically 1 room/4 pods/40 cabinets/250 devices; large DC 6/30/360/2000.
5. Example architecture aligns pod subsystems at 60 kW increments.
6. N+1 chiller plant: each chiller supports 4.5 pods; two on day one, a third at the fifth pod.
7. Large shared buses force up-front design of every configuration; independent per-pod buses scale best.

Using it in design

• Build the data model on the facility > room > pod > cabinet > device hierarchy and give each subsystem a deployment level and a linkage rule (for example 1 CRAH per N racks).
• Render a modularity chart of subsystems against IT kW in the phasing UI and verify N+1/2N redundancy at every phase.
• Warn when a shared parallel bus is sized to the final build while phase 1 is small (stranded switchgear and piping).
• Store PUE-versus-load, floor loading, cost per kW and kW per cabinet per architecture template so options can be compared before detailed design.

Who should read it

Planners documenting and comparing modular, phased data center architectures.

Related standards

• Uptime Institute Tier Standard: Topology
• TIA-942
• EN 50600-1

Publication details

Authors

Neil Rasmussen

Publisher

Schneider Electric

Year

2017

Document number

WP 160

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