Improve Data Center Cooling Facility Efficiency through Platform Power Telemetry (OCP DCF white paper, Rev 1.0)

Overview

The paper was written by Intel for the Open Compute Project Data Center Facility group (Rev 1.0, 2019). It addresses the use of server, rack and PDU power telemetry in DCIM to control facility cooling. The paper aims to show the benefit of a common telemetry approach in environments where the facility and IT sides are usually managed separately.

The first part describes a power telemetry hierarchy running from facility power distribution to the rack management controller, the server node and processor and memory sensors. Time-series design choices such as sampling period and change-triggered reporting are discussed. In an analysis of server power changes, more than 99.7% of change intervals were 20 seconds or longer, which guides telemetry sampling granularity (500 ms-1 s polling is compared with minute-level DCIM sampling).

The second part describes a predictive cooling control scheme. When forecast IT power is fed into the control loop, cooling responds earlier and overshoot is reduced, which shrinks the cooling buffer that would otherwise be overprovisioned. The method aims to bring cooling capacity forward in order to reduce overprovisioning. This approach lets cooling act on expected load rather than only reacting to measured temperature, especially in environments where load changes quickly.

The third part recommends DMTF Redfish (RESTful/JSON) to unify IT and facility telemetry. Facility protocols such as Modbus, BACnet, SNMP and IPMI are bridged into a single DCIM telemetry model through protocol proxies. The paper is a vendor study, and its results rest on the authors' scenarios. Such a unified model provides the foundation needed to gather facility and IT data on the same dashboard and feed it to control algorithms.

Key points

1. The telemetry architecture spans facility power distribution, rack management controller, server node and CPU/memory sensors.
2. More than 99.7% of server power change intervals were 20 s or longer.
3. Sampling detail: 500 ms-1 s polling is compared with minute-level DCIM sampling.
4. Feeding predicted IT power into cooling control gives an earlier response and less overshoot.
5. This approach reduces the overprovisioned cooling buffer.
6. DMTF Redfish (RESTful/JSON) with protocol proxies bridges Modbus, BACnet, SNMP and IPMI into one DCIM model.
7. The paper is Intel's 2019 work, with all rights reserved.

Using it in design

• In the monitoring/BMS design module, include a telemetry point list that pulls IT power (Redfish/IPMI) alongside facility meters.
• Keep sampling intervals at 20 s or less for control use.
• Offer predictive cooling control (IT power forecast to CRAH/CDU setpoints) as an optional controls-design feature.
• List Redfish and protocol gateways in DCIM integration specifications.
• When sizing the cooling buffer, assess how a forecast IT power profile can reduce overprovisioning.

Who should read it

Controls engineers designing monitoring, BMS and DCIM, and facility teams working on cooling efficiency.

Related standards

• DMTF Redfish (DCIM schema)
• IPMI 2.0
• ASHRAE 135 BACnet

Publication details

Authors

Open Compute Project Foundation

Publisher

Open Compute Project Foundation

Year

2019

Document number

Rev 1.0

Document language

English

Source type

Industry body

Access

Free to read

Licence

Bibliographic reference only — the document is not redistributed here; confirm reuse terms at the original source

Citation