Airflow distribution through perforated tiles in close proximity to computer room air-Conditioning units.

INTRODUCTION

Satisfactory thermal management of high power density data centers depends largely on the efficiency of delivering cool supply air from the computer room air-Conditioning units (CRACs) to the front of the data processing racks. Most data centers in existence today house the CRAC units inside the data center facility and utilize a raised floor plenum (RFP) to distribute cool supply air through perforated tiles located near the racks (ASHRAE 2005) and the hot exhaust air is collected from the upper portion of the facility. The racks are usually arranged in alternating directions of hot exhaust air, forming "hot" and "cold" aisles, in order to maximize the supply and return efficiency to the CRAC units (ASHRAE 2005) as shown in Figure 1. The RFP is also used for routing chilled-water supply and return piping to the CRAC units, large bundles of data processing cables, and power distribution equipment. These items may introduce significant blockage effects, severely altering the plenum pressure distribution and adversely affecting the flow distribution through the perforated tiles. The plenum footprint, depth, CRAC position, perforated tile porosity, and number of perforated tiles are also significant parameters in data center design (Rambo and Joshi 2006).

[FIGURE 1 OMITTED]

The objective of this paper is to highlight the differences that occur in perforated tile flow rate and flow distribution for tiles located in close proximity to CRAC units for RFP data centers with deeper plena compared to the existing measurement database for relatively shallow plena. Detailed pointwise pressure and airflow velocity measurements have not yet been performed in the plenum, but the presentation of the perforated tile flow rate is an important contribution to RFP data center design and diagnostics because is extends the very limited experimental database to deeper plena that are becoming increasing more common in practice.

This investigation presents an experimental investigation of plenum airflow near the CRAC unit, with special attention to reversed flow, or warm air from the facility being drawn down into the RFP due to a relatively low static pressure in the plenum. Reversed flow was experimentally observed by Schmidt et al. (2001) in a 0.284 m (11 in.) deep plenum for tiles located near the CRAC unit. A two-dimensional (depth-averaged) computational model was developed that predicts the general tile flow rate trends but fails to capture sharp variations between neighboring tiles and in some cases does not resolve experimentally measured reversed flow. A fully three-dimensional model of the same facility (Schmidt et al. 2004) showed improved agreement with experimental measurements but also failed to predict the reversed flow in many tile configurations. The experimental data of Schmidt et al. (2001, 2004) are from a data center with one CRAC unit near an artificially created wall and the other unit located 15 tiles away in the middle of the plenum. Some of the data indicate use of turning vanes to direct the CRAC unit exhaust in various directions. The term exhaust refers to the discharge of the CRAC unit into the RFP and does not imply any release of air to the ambient because data centers are closed-loop systems for all practical purposes.

With both units operating, the data show that the perforated tile located directly next to an operational CRAC unit will experience reversed flow. Tiles located up to four places away from the CRAC unit experienced reversed flow of negligibly small flow rates with the other unit shut off, in agreement with the asymptotic modeling presented in Rambo and Joshi (2004). This experimental data center has been used for other measurements and validation of computational models; however, the results are limited by a relatively shallow plenum (<0.30 m, ~12 in.) and a unique CRAC layout. Most operational RFP data centers locate the CRAC unit near the periphery of the facility, as characterized by Schmidt and Iyengar (2005) and Sharma et al. (2004) and computationally modeled by Van Gilder and Schmidt (2005).

Radmehr et al. (2005) experimentally investigated the leakage flow, or the portion of total CRAC airflow that seeps through the seams of the raised floor tiles, in a 0.42 m (16.5 in.) deep facility. Distributing the leakage flow uniformly throughout the perforated tile seams and modeling chilled-water supply lines, Radmehr et al. (2005) obtained predictions with an overall accuracy of 90%. Van Gilder and Schmidt (2005) present a parametric study of plenum airflow for various data center footprints, tile arrangements, tile porosity, and plenum depth. A vacant plenum was assumed, making the perforated tile resistance much greater than any other flow resistance in the plenum and allowing the resulting perforated tile flow rate to scale with the net CRAC flow rate.

The above investigations are, to the best of the author's knowledge, the only published experimental data and comparisons with numerical models for RFP data centers that can be used to address the issue of reversed or reduced flow rate though perforated tiles. Figure 2a-2g graphically illustrate the data in Schmidt et al. (2001, 2004), VanGilder and Schmidt (2005), and Radmehr et al. (2005), and Table 1summarizes the results. The figures shows the data center layout, CRAC location, and perforated tile arrangement. The "reduced" or "low" flow rate condition has been defined here as a perforated tile flow rate with less than 50 cfm, which represents 1 kW of cooling capacity at 35[degrees]C [DELTA]T for the data processing equipment. An "X" indicates the numerical model prediction error is either greater than 20% or 50 cfm. The 50 cfm criterion is used in the reduced flow rate cases, where 20% error may only account for a flow rate of a few cfm and the model prediction is sufficiently accurate. Table 1 summarizes the previous RFP data center perforated tile measurements and the corresponding modeling efforts.

[FIGURE 2 OMITTED]

The general conclusions that can be drawn from existing literature is that a perforated tile located adjacent to or one position away from an operating CRAC unit will experience reversed flow (see Figure 2a-2f and 2i-2j). An exception occurs if there are few enough tiles to cause a large enough static pressure in the RFP near the CRAC exhaust, as is shown in Figure 2g-2h and Figure 2n-2o. A scoop eliminates reversed flow (see Figure 2d and 2i. The two-dimensional numerical models of Figure 2a-2c show significant deviations from the measured values near operating CRAC units and in the shear layer where the two CRAC exhausts encounter one another. Results of the three-dimensional model in Figure 2e- 2f show the same error in reversed and low flow rate tile prediction as well as some additional error near the non-operating CRAC unit of Figure 2f. Also note it is believed that the data in Figure 2a-2o is for the same facility, although the authors (Schmidt et al. 2001, 2004; VanGilder and Schmidt 2005; Radmehr et al. 2005) report various plenum depths. Another shortcoming of the previously reported results is that no statistical information regarding the data collection processes is provided, and the variation in perforated tile flow rate is unknown.

Table 1. Summary of Previous RFP Perforated Tile Measurements

Author         Year  RFP Depth, m   Tile %  Model    Code    Figure
                      (in.)         Open
                                    Area

Schmidt et al. 2001  0.284 (11.2)     19%     2D   COMPACT    2a-2c


Schmidt et al. 2004  0.292 (11.5)  19.50%     3D   TileFlow   2d-2l*

VanGilder      2005  0.292 (11.5)  19.50%     3D   Flovent       2m
and
Schmidt

Radmehr et al. 2005  0.419 (16.5)     16%     3D   TileFlow   2n-2o


* Data reported for cases (e) and (f) only.

EXPERIMENTAL FACILITY LAYOUT

All measurements were performed at the 104 [m.sup.2] (1120 f[t.sup.2]) experimental data center laboratory at the Georgia Institute of Technology, constructed based on the four rows of seven racks unit cell architecture (Rambo and Joshi 2003). The facility contains two CRAC units capable of supplying the RFP ("downflow") and an additional two units that supply an overhead cooling system ("upflow") not incorporated in this investigation. One CRAC unit is denoted "A" for APC series CCT-CW, with a nominal flow rate of 6.090 kg/s at 15[degrees]C (10,500 cfm at 59[degrees]F), and the other is denoted "L" for the Liebert series FH529C with nominal flow rate of 7.169 kg/s at 15[degrees]C (12,400 cfm at 59[degrees]F). Both CRAC units drive the flow with two blowers in nearly identical layouts, and both units use the same blower, making the exhaust flow characteristics of the units very similar, with the small change in flow rate due to different internal pressure drops. CRAC-L is equipped with a variable-speed blower that varies the unit's flow rate depending on the heat load in the experimental facility. The laboratory is vacant of all computing equipment, causing the CRAC unit to output the minimum airflow rate, which is limited to 60% of the maximum to protect unit. The maximum CRAC unit flow rate is obtained by prescribing the temperature setpoint well below the ambient temperature and turning off the chilled-water supply. The blower units in CRAC-A are attached to a fixed-speed drive and its flow rate cannot be controlled. Neither CRAC unit employs a turning vane.

The data center raised floor is constructed of 0.61 m (24 in.) square tiles and is supported by an array of 0.025 m (1 in.) square stanchions located at the corners of the floor tile. The RFP measures 0.86 m (34 in.) to the bottom of the tiles, which is significantly deeper than other reported experimental results for RFP data centers. Figure 3 shows the blockages in the plenum, including the vertical location of the insulated chilled-water supply and return to CRAC units, drain piping, and cable trays. All piping was mounted as close to the raised floor as possible during the construction of the experimental facility to later alter the plenum depth. The facility also contains 15 power cables with nominal ~0.04 m (1.5 in.) diameter. Figure 4a shows photographs of the chilled-water supply piping and Figure 4b shows the stanchions, piping, and electrical power cables.

[FIGURE 3 OMITTED]

[FIGURE 4 OMITTED]

The experimental facility contains 32 identical perforated tiles with 5.95 mm diameter holes arranged on a staggered pitch for an open area of 21.26%. Data center perforated tiles are nominally 25% open area, and this discrepancy has also been observed in Schmidt et al. (2001, 2004) and Radmehr et al. (2005). The perforated tiles are a combination of a 0.00635 m (1/4 in.) thick perforated screen attached to a 0.028575 m (1-1/8 in.) thick structural support with nominal open area of 30.68%. Figure 5 is top and bottom photographs of the perforated tiles.

[FIGURE 5 OMITTED]

The experimental measurements are performed with the 32 perforated tiles arranged to form two cold aisles 2 tiles wide by 8 tiles long. The perforated tiles occupy the y-locations (Y = {D, E, ..., K}, cf. Figure 3), and it is assumed that there are five solid tiles between the cold aisles, which provides a representative spacing between cold aisles considering standard rack dimensions. With the number and porosity of the perforated tiles and the tile configuration constrained as above, the distance to the first row of perforated tiles and operating status of the CRAC unit are the only sources of variability.

EXPERIMENTAL MEASUREMENTS

Individual tile mass flow rates were measured using a Shortridge Instruments AirData Multimeter ADM-850L attached to a Flowhood array of pitot static tubes specifically designed to measure airflow rates from 0.61 X 0.61 m (2 X 2 ft) ceiling and floor tiles. The instrument is capable of correcting for its own flow resistance, local temperature, and back-pressure with an accuracy of [+ or -]3% [+ or -]7 cfm as calibrated by the instrument manufacturer over a range of flow rates from 25 to 2000 cfm. The measurements were also corrected for humidity because the data were collected over a period of several days in Atlanta, Georgia, where the relative humidity varied between 40% and 60% in the experimental facility. Ten data points were collected for each tile flow rate in each configuration. Postprocessing the data showed that the standard deviation is an accurate estimate of 90% confidence interval based on the number of samples, measurement accuracy, and estimated standard deviation assuming normally distributed data. The data collection procedure assumes steady operation because only one tile flow rate can be measured at a time.

Due to the size and location of the CRAC units, the flow rate instrument cannot be used to measure the return flow rate, and the CRAC unit's exhaust flow rate exceeds the instrument's range. All exterior walls of the plenum were inspected for air leakage using a commercial smoke generation machine and a multi-directional hotwire anemometer, and all possible leakage points were sealed to ensure that all airflow from the CRAC units remains in the RFP. The hotwire anemometer probe was also used to verify that a negligibly small amount of air leaked into the facility through the seams between tiles.

EXPERIMENTAL RESULTS

Experimental configurations will be identified with the nomenclature corresponding to the CRAC unit(s) in operation and the location of the first row of perforated tiles. The nomenclature A denotes CRAC-A operating, L denotes CRAC-L operating, and AL for both CRAC units being on. Lr and ALr indicate CRAC-L is operating in the reduced flow rate mode, with CRAC-A off and on, respectively. The numerals X = {3, 4, 5} are used to denote the location of the first row of perforated tiles. For example, the configuration with the first row of tiles located at X = 3 and both CRAC-A and CRAC-L operating is designated 3AL. Individual tiles are identified by their alphanumeric positions on the tile grid in Figure 3.

It was argued in VanGilder and Schmidt (2005) that perforated tile flow distribution becomes independent of the net CRAC flow rate when the perforated tile resistance is much greater than any other resistance in the airflow path by taking the tile resistances in parallel. This may be true for perforated tiles located "far" from the CRAC units in the limit of no other significant obstructions in the plenum. This argument disregards the effect of perforated tile position relative to the CRAC units, which has a significant effect when the perforated tiles are in close proximity to the CRAC units and the individual exhaust jets produce more complex flow patterns. The argument that tile flow rate distribution scales directly with net CRAC flow rate implies reversed flow rate tiles would experience greater reversed flow rates as the net CRAC flow rate increased.

Figure 6 graphically depicts the mean measured perforated tile flow rates for all combinations of the CRAC units operating at full flow rate capacity. The immediate observation is that reversed flow only occurs at tile K3 for the 3L configuration and K4 for the 4L configuration. Low flow rate tiles only occur at location F3 for the configuration 3A and at locations J3 and K4 for configuration 3L. No reversed flow occurs with both CRAC units operating. Some of the reduced flow rate tiles have mean values less than 25 cfm, which is below the range of instrument. It should be acknowledged that the net flow rate through that specific tile is severely reduced relative to the other tiles in the facility and is somewhere in the range of [+ or -]25 cfm.

[FIGURE 6 OMITTED]

The experimental layout presented here is most similar to that of previous measurements in Figure 2b, 2f, and 2l because the operating CRAC unit is near a wall, forcing its exhaust in the direction of the perforated tiles. The scenarios in Figure 2c, 2d, and 2f exhibit some similarity because the other CRAC unit does not impede the exhaust's path to the perforated tiles. Reversed flow rates occur up to three tiles away from the CRAC unit, and low flow rates are observed up to five tiles from the front of the CRAC unit. Figure 2l shows reversed flow when the first row of perforated tiles is located five positions away from the front of the CRAC unit. The measurements of Figure 6 show significant variations from the previous results.

The first observation from the data is that common rows between two different configurations produce similar tile flow distributions, suggesting the flow distribution through a row of perforated tiles is relatively independent of the upstream (X decreasing) or downstream (X increasing) row being populated with solid or perforated tiles. The configurations 3A and 4A both have perforated tiles located at X = 4 and X = 11 and the configurations 4A and 5A both have perforated tiles located at X = 5 and X =12. This is a useful property because reasonable estimates for the flow distribution through the rows X = 4 and X =5 in the 4AL configuration can be obtained by using the data from row X = 4 from the 3AL case and row X = 5 from the 5AL case. Figure 7 compares the perforated tile flow rate response for all common perforated tile rows.

[FIGURE 7 OMITTED]

Another important shift in the perforated tiles occurs in the Y-direction, changing the relative position between the row of perforated tiles and the CRAC units. All the perforated tiles in Figure 2 are located directly in front of the CRAC units, although this may not be the case in an actual data center. Figure 8 illustrates the perforated flow rate for the configurations 3L and 3AL with the perforated tiles in row X = 3 and X = 4 shifted by three positions in the positive Y-direction. The tiles that experienced reversed flow rate (K3) and reduced flow rate (J3 and K4) exhibit the same behavior independent of the shift. The tiles in positions G through K have very similar flow rates for both original and shift Y-positions, indicating that shifting tiles in this direction does not significantly alter the flow rate and that the perforated tile's flow rate is relatively independent of whether the Y-direction neighbor is solid or perforated. Note the tiles located at X = 10 and X = 11 were also shifted, but the results were not reported as the same trends of rows X = 3 and X = 4 were observed.

[FIGURE 8 OMITTED]

With only chilled-water piping and cable trays in the plenum, the data center is designed to be much more similar to the "vacant" condition commonly modeled than an operational data center with large bundles of data and power cables restricting the flow. The facility schematic of Figure 3 shows the main chilled-water piping runs under tiles D to N in rows X = 5 and X = 6 and the CRAC unit feeder line runs under tiles D, E, J, and K from X = 2 to X = 5. Since the piping is mounted close to the raised floor (cf. Figures 3-4), one would expect to see a reduction in flow rate for perforated tiles located above the piping. However, the data in Figure 6 show no discernible effects of the piping.

To quantify the effect of blockage area, consider the projected area of all the possible obstructions under a single tile onto the x-z plane. The area of the stanchions provides a 4.16% blockage. Assuming that the obstructions run completely through the volume, the main water supply line blocks 16.28%, the feeder line blocks 11.86%, and the cable tray blocks an additional 4.65%. The drain line does not provide a blockage because it runs concurrently with the main supply line. If all these blockages were to occur in one region simultaneously, the flow would experience a local blockage area of 36.95%. No such situation occurs in this data center, and the maximum blockage ratio under a single tile projected onto the x-z plane is estimated to be ~ 21%, illustrating that significant blockages may occur for an RFP containing only the essential structure to support the raised floor and run the CRAC units.

It is difficult to determine whether decreased flow rate is caused by the tile layout, CRAC unit operation, or blockages in the plenum. Figure 2 shows chilled-water piping spanning across data center rows X = 5 and X = 6 and down column D. The tile flow rates for the second row in configuration X = 5 (row 6 in the data center) do not show any decreases in flow rate. The effect of the cable tray cannot be determined because it is located near the CRAC units (data center row 4 and down column F), and one cannot determine if low flow rate or reversed flow rates are a result of the close proximity to the CRAC units or a blockage effect from tile measurements alone. Since no cables are routed on the cable tray, making it mostly open area when viewed from above, and since it is relatively thin, one would expect its flow resistance to be of secondary importance, especially in the CRAC exhaust region.

To validate the claims of tile flow distribution being independent of CRAC flow rate (VanGilder and Schmidt 2005) measurements were performed for the configurations 3Lr, 5Lr, 3ALr, and 5ALr, where the "r" denotes the CRAC-L is operating at a reduced flow rate nominally equal to 60% of its total flow rate capacity. Figure 9 illustrates the mean flow rate for the reduced flow rate cases. Tile K3 exhibits reversed flow and tiles J3 and K4 show reduced flow for the 3Lr case, just as with the 3L case.

[FIGURE 9 OMITTED]

To better understand the effect of CRAC flow rate, the individual perforated tile flow rate for each case was normalized by the sum of all the mean perforated tile flow rates for that configuration. A ratio of unity for each tile between the normalized flow rates for the reduced and total flow rate cases would indicate that the flow rate distribution for that configuration is independent of the net CRAC flow rate. Figure 10 plots this ratio, and the vertical axis has been rescaled about 1[+ or -]0.2 to better highlight the variations.

[FIGURE 10 OMITTED]

The data in Figure 10 indicate that the tile flow distribution is generally dependent on CRAC flow rate. The experimental error is less than 10% so the normalized flow rate ratio outside the range 1 [+ or -] 0.1 deviates significantly enough to show that particular tile's flow rate for that particular configuration is dependent upon net CRAC flow rate. The data indicate that perforated tiles in close proximity to the CRAC units are strongly dependent upon on CRAC flow rate and there are some minor variations in perforated tiles far from the CRAC unit. The most significant deviations occur for the reversed and reduced flow tiles. Tiles J3, K3, and K4 for the 3L-3Lr comparison show greater than 20% variation. Generally, the departure from CRAC flow rate independence is greater for both tiles located closer to the CRAC units as well as the ALr configuration (when the CRAC-A is on full and CRAC-L is reduced). Perforated tile distribution dependence on CRAC flow rate is observed up to five tile positions away from the CRAC unit for tiles F5 and I6 for the 5L-5Lr comparison and for tiles H6 and I6 for the AL-ALr comparison. Tiles located at X = 10 through X = 13 did not exhibit a significant distribution dependence on CRAC flow rate.

Repeated measurements for each tile for each configuration were performed in order to collect statistically significant data. Some large ranges in the measured values were observed, which is reflected in the plots of standard deviation in Figures 11 and 12. Larger standard deviations are observed for tiles close to the CRAC units and the measurement variations decrease for tiles farther from the CRAC units. The largest variations between successive measurements was observed with CRAC-A operating only and then with both CRAC-A and CRAC-L operating, and the cases with CRAC-L operating only exhibited the smallest variations. The reduced flow rate cases have a lower variation in measured value as compared to full flow rate cases. In all the configurations, including the shifted cases of Figure 8, the large variations occurred for perforated tiles located in the first row of perforated tiles and directly in front of the operating CRAC units. Tile K3 in the 3L configuration exhibits a large standard deviation, as well as tile H3 in the 3AL configuration, indicating that tiles with both reversed and normal flow rates may show a considerable range of variability.

Without more detailed measurements, the cause of the large perforated tile flow variances cannot be determined directly. The deep plenum will cause stronger pressure variations in the vertical direction, especially near the CRAC exhaust units. Previous work has reported the average static pressure in the RFP (Radmehr et al. 2005; Koplin 2003) or two-dimensional maps of static pressure measurements (Koplin 2003) without indicating any variations in the vertical direction. The variations could possibly be to due to turbulent intermittency, a dynamic output from the CRAC exhaust or an effect of the flow obstructions in the RFP.

[FIGURE 11 OMITTED]

[FIGURE 12 OMITTED]

CONCLUSIONS

Experimental measurements were performed in an experimental RFP center with 0.86 m deep plenum and two downflow CRAC units located next to each other. The results showed significant deviations from tile flow rate trends derived from other experimental measurements and computational results for less deep plena without the minor blockages. The significant trends for the data collected are as follows:

1. Diminished, but not reversed, flow rate was observed for most perforated tiles located adjacent or one tile position away from the CRAC unit.

2. Reversed flow occurred in a few perforated tiles for only a specific combination of CRAC operation and does not require the perforated tile to be placed directly in front of an operating CRAC unit.

3. Near the CRAC units, tile flow rate was found to depend primarily on position, with low or reversed flow rates occurring relatively independently of neighboring tiles.

4. Chilled-water piping mounted near perforated tiles did not have any significant effect on the flow rate of tiles located directly above relative to other neighboring perforated tiles.

These conclusions are valid only for the data presented here, but they highlight differences in tile flow rates for a deeper plenum than previously measured. They also suggest that the flow in the plenum, especially in the CRAC exhaust region, is very complex, and more advanced modeling techniques may be required to fully understand the flow rate distribution through perforated tiles in raised floor plenum data centers. This is the first report of both perforated tile flow rate data for a deep plenum data center and for two neighboring downflow CRAC units, as commonly found in operational data centers.

ACKNOWLEDGMENTS

The authors acknowledge support of this work through the Consortium for Energy Efficient Thermal Management (CEETherm), a joint initiative between Georgia Institute of Technology, the University of Maryland, and industry partners.

REFERENCES

ASHRAE. 2005. Datacom Equipment Power Trends and Cooling Applications. Atlanta: American Society of Heating, Refrigerating and Air-Conditioning Engineers, Inc.

Koplin, E.C. 2003. Data center cooling. ASHRAE Journal 45(3):46-53.

Radmehr, A., R.R. Schmidt, K.C. Karki, and S.V. Patankar. 2005. Distributed leakage flow in raised-floor data centers. Proc. of IPACK'05, International Electronic Packaging Technical Conference and Exhibition, San Franciso, CA, July 17-22.

Rambo, J., and Y. Joshi. 2006. Convective transport processes in data centers. Numerical Heat Transfer A-- Applications 49(10):923-45.

Rambo, J., and Y. Joshi. 2004. Supply air distribution from a single air handling unit in a raised floor plenum data center. Proc. of ISHMT/ASME'04, Joint Indian Society of Heat and Mass Transfer-American Society of Mechanical Engineers Heat and Mass Transfer Conference, Kalpakkam, India, January 5-7.

Rambo, J., and Y. Joshi. 2003. Physical models in data center airflow simulations. Proc. of IMECE'03, ASME International Mechanical Engineering Congress and R&D Exposition, Washington DC, November 16-21.

Schmidt, R.R., K.C. Karki, K.M. Kelkar, A. Radmehr, and S.V. Patankar. 2001. Measurements and predictions of the flow distribution through perforated tiles in raised floor data centers. Proc. of IPACK'01, The Pacific Rim/ ASME International Electronics Packaging Technical Conference and Exhibition, Kauai, HI, July 8-13.

Schmidt, R.R., K.C. Karki, and S.V. Patankar. 2004. Raisedfloor data center: Perforated tile flow rates for various tile layouts. Proc. of ITHERM 2004, Ninth Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems, Las Vegas, NV, May 4. pp. 571-78.

Schmidt, R.R., and M. Iyengar. 2005. Effect of data center layout on rack inlet air temperatures. Proc. of IPACK'05, International Electronic Packaging Technical Conference and Exhibition, San Francisco, CA, July 17-22.

Sharma, R., C.E. Bash, C.D. Patel, and M. Beitelmal. 2004. Experimental investigation of design and performance of data centers. Proc. of ITHERM 2004, Ninth Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems, Las Vegas, NV, May 4. pp. 579-85.

VanGilder, J.W., and R.R. Schmidt. 2005. Airflow uniformity through perforated tiles in a raised-floor data center. Proc. of IPACK'05, International Electronic Packaging Technical Conference and Exhibition, San Francisco, CA, July 17-22.

Jeffrey Rambo and Graham Nelson are graduate research assistants and Yogendra Joshi is a professor and associate chair for graduate studies at the Georgia Institute of Technology, Atlanta, GA. Dr. Joshi holds the John M. McKenney and Warren D. Shiver Distinguished Chair in Building Mechanical Systems.

Jeffrey Rambo

Graham Nelson

Yogendra Joshi, PhD

Member ASHRAE