Sign In
Forgot Password?
Sign In | | Create Account

11 Top Tips for Energy-Efficient Data Center Design and Operation… A High-Level ‘How To’ Guide

White Paper When is Data Center Thermal Design Important? The usage of data centers globally has skyrocketed in the past three to four years in response to growing demand for information storage and transfer from banks,...

Engineering Edge

Implementing Practical Fan Curves in Datacenter Simulations

Datacenters are among the most energy consuming facilities. The current trend is continuously increasing to satisfy the growing needs of E-commerce and other revolutionary technologies such as cloud computing. We use CFD, to model variable flow devices that exist in a typical datacenter.

By Sami Alkharabsheh, State University of New York at Binghamton

Mechanical Analysis Division’s FloVENT® product has been the simulation engine for a datacenter modeling for 25 years. In a new track at the IEEE SEMI-THERM conference in 2013 a technical paper presented [1] that investigated the validity of a common modeling assumption when using Computational Fluid Dynamics (CFD) to predict and analyze air flow and temperatures in datacenters.

It is often assumed by CFD practitioners when performing datacenter simulations that the volumetric air flow rates provided by a Computer Room Air Conditioning (CRAC) unit and a server are constants. The rationale for this approach is the assumption that the resistance to air flow within a CRAC unit and a server will be far larger than any other air flow resistance present in the datacenter. If that’s true, then the system impedance curve of the CRAC unit can be used with the CRAC blower performance curve to read off the operation point of the unit. A similar approach can also be applied to servers. Figure 1 shows the representative datacenter considered in this study.

Figure 1 Plan view of the datacenter layout being considered

The internal elements of CRAC units include filters, cooling coils, baffles, ducting, and is generally a very cluttered and restrictive piece of equipment to move air through. The other sources of pressure drop in the datacenter are all external to the CRAC unit: movement of air throughout the datacenter space, under the raised floor, through the perforated tiles and in and around the electronics equipment. The assumption investigated in our paper was how large is the ratio of ‘external’ to ‘internal’ losses in a datacenter, and what impact could this have on the commonly used fixed flow rate in the modeling method of the CRAC units and the servers. The chart in Figures 2 & 3 demonstrates the potential impact on the expected flow rate from the CRAC unit blower if the ‘external’ losses are 10% and 25% of the ‘internal’ losses.

Figure 2 Blower curve used with a typical system impedance curve

Figure 3 Effect of the external static pressure on the operating point of the CRAC blower

The change in CRAC unit flow rate by considering the external losses are indicated by the horizontal arrows. For this chart, the impact of including the external loss in the impedance curve is not large as the blower performance curve has a steep slope in the vicinity of the operating point. If the operating point was near a flatter portion of the blower curve, then the CRAC flow rate would be more sensitive to changes in the external static pressure contributions that would be caused by adding or removing electronics equipment, partitions or floor tiles.

With an application specific simulation tool like FloVENT we can investigate such assumptions in real environments as was done in the paper. FloVENT has SmartParts for various objects commonly needed for a datacenter analysis, such as CRAC units, racks, perforated tiles, fan curves, flow resistances, as well as a parts library for easy retrieval of previously defined parts. Each of these SmartPart objects has a specific set of tailored data available after a solution. For a fan curve definition, this tailored data includes the operating point of the fan, and thus the static pressure observed by the fan in the model.

The methodology[1] to evaluate the CRAC fixed flow rate assumption involves the creation of a datacenter model and using the object specific results reporting in FloVENT to extract the external system impedance for each CRAC unit, followed by calibration of the internal flow resistance definition to match the observed flow rate with the manufacturer’s blower performance curve. For the final calibrated operating point, in the paper[1], the external static pressure was 0.22 inches of H2O and the internal static pressure was 2.82 inches of H20, a ratio for external/internal of 7.8%.

The CRAC unit models with calibrated fan curves were placed in a simulation for an approximately 1,723 ft2 datacenter with a typical hot aisle, cold aisle rack layout. The variation in flow rates from the calibrated fan curve CRAC models was less than 120 ft3/min, which is less than 1% of the average CRAC unit flow rate. This data suggests that the common fixed flow rate assumption for CRAC units would produce satisfactory results in this model. This was the expected result with the datacenter model in question being considered fairly open. It was suggested in reference 1 that future work from the authors could explore the sensitivity of this approach in datacenters where the external impedance curve is expected to be more significant, such as in containment systems.

Figure 4 Plan view of the datacenter layout being considered with FloVENT predicted temperature contours

Reference:

  1. S. Alkharabsheh, B. Sammakia, S. Shrivastava, R. Schmidt, “Utilizing Practical Fan Curves in CFD Modeling of a Data Center”, SEMI-THERM 2013, pp 208-212.
 
Online Chat