Airflow uniformity through perforated tiles in a raised-floor data center

Airflow Uniformity through Perforated Tiles in a Raised-Floor Data Center by James W. VanGilder, P.E., APC by Schneider Electric Roger R. Schmidt, Ph.D., P.E., IBM Corporation Executive summary Perforated tiles on a raised floor often deliver sub-stantially more or less airflow than expected, resulting in inefficiencies and even equipment failure due to inadequate cooling. In this paper, the impact of data center design parameters on perforated tile airflow is quantified and methods of improving airflow unifor-mity are discussed. This paper was written jointly by APC and IBM for the ASME InterPACK '05 conference. This white paper was originally published in Proceedings of IPACK2005. ASME InterPACK '05, July 17-22, 2005 San Francisco, California, USA white papers are now part of the Schneider Electric white paper library produced by Schneider Electric's Proceedings of IPACK2005
ASME InterPACK '05
July 17-22, San Francisco, California, USA
James W. VanGilder, P.E.
Roger R. Schmidt, Ph.D., P.E.
American Power Conversion Corporation 85 Rangeway Road Billerica, Massachusetts 01862 Poughkeepsie, New York 12601 [email protected] [email protected]
(CRAC's) pump conditioned air into the plenum; the air exits the plenum through perforated floor tiles and other openings. The maximum equipment power density (e.g. in Hot equipment exhaust air may be returned to the CRAC's power/rack or power/area) that may be deployed in a typical "through the room" or through an overhead return plenum or raised-floor data center is limited by perforated tile airflow. In ductwork. Emerging standard practice is to arrange the the design of a data center cooling system, a simple estimate of perforated floor tiles in rows forming a "cold aisle" (for mean airflow per perforated tile is typically made based on the example, see [1], [2], or [3]). Equipment racks are then placed number of CRAC's and number of perforated tiles (and in rows along and facing each long side of the cold aisle. possibly a leakage airflow estimate). However, in practice, Alternating hot and cold aisles are formed as this configuration many perforated tiles may deliver substantially more or less is repeated across the data center. than the mean, resulting in, at best, inefficiencies and, at worst, equipment failure due to inadequate cooling. Consequently, The temperature of the cooling air actually available for IT the data center designer needs to estimate the magnitude of equipment depends on the airflow dynamics between the variations in perforated tile airflow prior to construction or perforated tile and the equipment inlet. Equipment will draw air as needed and, if sufficient cooling air is unavailable, warm exhaust air will be recirculated over the racks or around the row In this paper, over 240 CFD models are analyzed to ends. It is, therefore, essential that perforated tiles located near determine the impact of data-center design parameters on the equipment provide sufficient cooling air. perforated tile airflow uniformity. The CFD models are based on actual data center floor plans and the CFD model is verified In some cases, a non-uniform airflow distribution through by comparison to experimental test data. the perforated tiles may be desirable. This may be achieved by using varying tile types or dampers. Techniques have been Perforated tile type and the presence of plenum developed to aid the design of such a system ([4] and [5]). The obstructions have the greatest potential influence on airflow present study is applicable to the majority of data centers, uniformity. Floor plan, plenum depth, and airflow leakage rate which are designed for uniform airflow through all tiles. have modest effect on uniformity and total airflow rate (or Results are equally valid for isolated-plenum zones, which may average plenum pressure) has virtually no effect. Good be present within a larger data center. For such uniform- uniformity may be realized by using more restrictive (e.g. 25%- airflow designs, the mean airflow per tile is usually estimated in open) perforated tiles, minimizing obstructions and leakage the design stage based on total CRAC airflow, the number of airflow, using deeper plenums, and using rectangular floor perforated tiles, and possibly an estimate of total leakage plans with standard hot aisle/cold aisle arrangements. airflow. In actual facilities, the airflow delivered through any particular perforated tile may vary substantially from the mean resulting in local cooling capacity, which is excessive or insufficient for the design IT load. Data centers contain servers and other IT equipment, which require precisely conditioned cooling air for predictable Thus, the data center designer needs to understand the performance. For example, ASHRAE's Thermal Guidelines parameters that affect airflow uniformity and the magnitude of for Data Processing Environments [1] recommends inlet influence of each. In this study, uniformity is assessed based temperatures in the range of 20-25∞C (68-77∞F) for Class 1 on percentage variation from the mean perforated tile airflow. (high-end, mission critical) equipment. The cooling air is For convenience, plenum geometry, CRAC details and typically distributed throughout the data center through a locations, perforated tile locations, and all other attributes raised-floor plenum. Computer Room Air Conditioners related to the basic layout of the facility are lumped into a Copyright 2005 by ASME category that will simply be called "floor plan". With this Uniformity does not become perfect, as the plenum is definition, the parameters investigated here are: 1) floor plan 2) made very deep. While this would be true if air were perforated tile type (% open area), 3) airflow leakage rate, and introduced from the bottom of the plenum, in real data centers, 4) plenum depth. CRAC's supply air from the top of the plenum - just under the raised floor. Consequently, substantial air movement takes Ten floor plans are studied; one is a hypothetical layout, place directly under the tiles. Jets and recirculation zones are nine are based on actual data centers. Of the nine actual data characterized by regions of rapidly varying air velocity which centers, one is a facility for which measurement data has been may sustain pressure variations on the order of ½ r V2. taken; this case is used to validate the CFD model. Two Perforated tiles must offer substantially more resistance (greater perforated tile types are considered: nominally 25%-open and pressure drop) than these pressure fluctuations to ensure nominally 56%-open tiles. Leakage rate is the airflow rate uniformity. Assuming the velocity of air through the tiles and associated with holes in the raised floor for cable access and that in the plenum are of the same scale, we have ftile>>1 as a other openings generally not intended primarily for air delivery. minimum requirement for uniformity. In practice, these flow- Three leakage rates are considered in the range of 0% to 40% induced pressure variations may not be the dominant resistance of total CRAC airflow. Four plenum depths in the range of in the plenum; bounding walls, obstructions (e.g. cable trays, 0.30 m (12 in) to 0.91 m (36 in) are considered. piping, stanchions, etc.), and pressure loss through leakage through non-perforated-tile openings may dominate and the Airflow rate (e.g. total from the CRAC's or mean airflow perforated tiles must be even more restrictive in order for per perforated tile) and average plenum pressure are notably reasonable uniformity to be achieved. absent from the list of parameters studied. Schmidt et al [6] suggested that since all relevant pressure variations scale with Plenum pressure may drop below room pressure in certain the velocity squared, airflow patterns in the plenum would regions so that "backflow" occurs through perforated tiles. remain independent of airflow rate. Accordingly, if CRAC Similarly, "leakage" airflow through cable cutouts, around tile airflow rate were consistently scaled up or down there would be edges, and other openings may actually be directed into the no change in the percentage airflow variation from tile to tile. plenum. This leakage airflow may vary considerably over the The present study affirms this assertion and provides more raised floor but it locally varies approximately with the square discussion and examples. root of the pressure difference across the floor like the perforated tile airflow-pressure relationship expressed in Eq. Though not included in the CFD models of the present Perforated tile loss coefficient Airflow path loss coefficient investigation, obstructions such as cable trays and piping may Total number of perforated tiles in floor plan have a significant affect on perforated tile airflow in a particular region. Probability density as a function of Qi Airflow Uniformity is Independent of Airflow Rate Percentage variation from mean airflow rate for perforated tile i Standard deviation PLENUM AIRFLOW DYNAMICS
Uniform airflow through each perforated tile is achieved when the resistance to flow experienced by the air moving through the plenum is much less than the resistance to flow imposed by the perforated tiles; in this case, the total flow resistance along any path is fairly constant and no airflow path looks particularly attractive or unattractive. Both the airflow path resistance and the tile resistance scale approximately with the velocity (or flow rate) squared. The perforated tile resistance may then be characterized in terms of a loss As mentioned above, the fraction of total CRAC airflow through any perforated tile is largely independent of total where DP is the pressure drop across the perforated tile, r is the airflow rate and, therefore, average plenum pressure. This is density of air, and V is the velocity of the air approaching the fortuitous; we do not have to include airflow rate in our list of perforated tile. parameters considered and by expressing perforated tile airflow as a percentage (or fraction) of the mean per-tile airflow, we Copyright 2005 by ASME can readily compare results of scenarios with substantially differing airflow rates in a meaningful way. PERFORATED TILE AIRFLOW UNIFORMITY METRICS
For example purposes, consider a simple two-perforated- Airflow uniformity results are presented as percentage tile, single-CRAC floor plan. With reference to Fig. 1, an variations from the mean so that they may be readily applied to analogy may be made between airflow and the flow of current any airflow rate of interest and scenarios with different airflow in an electrical circuit. Assuming the airflow-path pressure rates may be compared on a consistent basis. The specific drop and perforated-tile pressure drop each scale with the flow uniformity metrics presented here are minimum, maximum, and rate (or velocity) squared, both sources of resistance may be standard deviation of the mean perforated tile airflow. A lumped into a single resistance. Each combined resistance, R1 negative percentage variation implies flow less than the mean. or R2, represents the total pressure drop associated with airflow A value less than –100% implies "backflow" from the room following a closed circuit starting at the CRAC, traveling down into the plenum. A positive percentage variation implies through the plenum, passing through one of the perforated tiles, flow greater than the mean. A value greater than 100% implies and ultimately returning to the CRAC. The total pressure drop flow that is more than twice the mean. As a concrete example, across either path is equal to the external pressure drop consider a scenario for which the mean tile airflow is 400 cfm overcome by the CRAC: with uniformity results reported as –75%, 50%, and 25% for the minimum, maximum, and standard deviation, respectively. In this case, at least one tile has the minimum airflow value of 100 cfm, at least one tile has the maximum airflow rate of 600 where q1 and q2 are the airflow rates along the two paths and k1 cfm, and the standard deviation is 100 cfm. and k2 are constants which characterize the combined path and tile resistances. Note that if k1=k2, a uniform distribution of Since the mean of the percentage variation from the mean airflow is achieved. It is also evident from Fig. 1 that: is zero by definition, the standard deviation may be written Solving Eqs. (2) and (3) simultaneously for q 2/QCRAC leads to: 1/QCRAC = k2 /(k1 + k2 ) (4a) i is the percentage variation in airflow from the mean for tile i, and n is the total number of perforated tiles in the floor plan. The probability density P(Q 2/QCRAC = k1 /(k1 + k2 ) (4b) i) is then (see [7] for Equations (4a) and (4b) show that the fraction of total airflow along each path is independent of airflow rate. Note that the above argument is also equally valid for floor plans using mixed tile types or tiles with dampers. The only requirement is that all pressure losses along each airflow path scale with the Equation (6) represents the classic bell curve that indicates the airflow rate (or velocity) squared. spread of variations from the mean. Assuming that this normal distribution implied by Eq. (6) applies adequately, 68% of all Generalizing, it can be shown that airflow uniformity is perforated tiles will have airflow within ±1s and 96% will be independent of airflow rate for any number of perforated tiles. Further, the pressure losses need not scale with airflow rate squared; the only requirement for uniformity is that all pressure PARAMETERS CONSIDERED IN CFD MODELS
losses scale identically (e.g. linearly, to the 1.5 power, etc.) with airflow. In practice, the airflow in a plenum is typically in As indicated above, the parameters investigated here are: the turbulent (high Reynolds Number) regime so that all 1) floor plan 2) tile type (% open area), 3) leakage rate, and 4) pressure losses will generally scale approximately with the airflow rate squared. Still it is conceivable that in some scenarios, different airflow regimes (laminar, turbulent, mixed) Tables 1 and 2 summarize the ten floor plans considered. may be present simultaneously. In this case, plenum and Note that the gridlines in the figures of Table 1 correspond to perforated tile losses will not scale identically with airflow rate standard 0.61 m (2 ft) x 0.61 m (2 ft) floor tiles and can be used and airflow uniformity will depend to some degree on the for scaling dimensions as desired. Floor Plan A is a magnitude of airflow rate. hypothetical floor plan; it follows the standard hot-aisle/cold- aisle layout with CRAC's located at the end of the hot aisles. CFD models were created to verify and demonstrate that As shown in the figure, mirror symmetry conditions are applied airflow uniformity is indeed independent of airflow rate for along the top and bottom walls in the CFD study so that the selected practical cases. Results are provided below. effective floor plan repeats indefinitely in both directions. Floor Plan A is also used for grid independence verification and for a cold-aisle-length investigation discussed below. Floor Plan B corresponds to an actual data center in which airflow Copyright 2005 by ASME

measurements were made; these results are used to verify the CFD model used for all scenarios. Floor Plans C through J are based on actual data centers, which have been previously Total Flo
ated ile Me
r low # of Per
40% leakage
modeled with CFD. Only the basic floor plans including 3 s) (cfm)
3 s) (cfm)
CRAC, perforated floor tile, and plenum-ducted IT equipment locations are retained for the present study. Two common perforated tile types are considered: nominally 25%-open and nominally 56%-open areas. As shown in Fig. 2, more-restrictive (e.g. 25% open) tiles typically feature drilled holes while high-flow tiles (e.g. 56% open) may more accurately be described as grates. Each scenario modeled features either all 25%-open or all 56%-open perforated tiles; no mixed-tile scenarios were considered. Based on manufacturer's published data, the loss coefficients (as defined in Eq. (1)) for the 25% and 56% perforated tiles are taken as 51.3 and 3.4 respectively. Note that some of the mean perforated tile airflow rates listed in Table 2 are unrealistically high for 25%-open tiles. Recall that the airflow uniformity results hold equally well at lower airflow rates. The airflow rates listed are the values used in the CFD models and are of value primarily for reference. Three leakage rates are considered: 0% (no leakage), 20%, and 40%. These leakage rates are defined as a percentage of total CRAC supply airflow minus any plenum-ducted IT airflow. These leakage rates are probably conservative in light of anecdotal evidence indicating that it is not uncommon for 50% or more of the airflow supplied by the CRAC's to exit the plenum via non-perforated-tile openings. Four plenum depths are considered which cover the range of typical design values: 0.30 m (12 in), 0.46 m (18 in), 0.61 m (24 in), and 0.91 m (36 in). The combination of design parameters discussed above results in 240 scenarios. Additionally, a supplemental floor- plan study is carried out using Floor Plan A with a 0.61 m (24 in) plenum and 20% leakage. The effect of row length on airflow uniformity is investigated; the room is made wider and narrower by adding and removing perforated tiles while keeping all other geometry and conditions fixed. Copyright 2005 by ASME models. It is noted that for floor plans including plenum-ducted CFD MODELING CONSIDERATIONS
IT equipment (E, G, and H), airflow uniformity results hold at different flow rates only when IT airflow is scaled up or down Only the plenum airflow is modeled in this investigation in proportion with the total CRAC and leakage airflow. with a zero pressure boundary condition imposed above the raised floor. In practice, the flow in the plenum is fairly FLOVENT V5.1 by Flomerics [8] was used for all CFD decoupled from the room flow for fairly restrictive perforated modeling. A structured Cartesian grid and the k-ε turbulence tiles like the 25%-open type. With less restrictive tiles, like the model were used for all simulations. Total simulation time for 56%-open type, the plenum and room (above the raised floor) the primary 240 scenarios was approximately 1 week on a 3.4 airflow are somewhat coupled. In this case, room airflow GHz Pentium 4 computer. dynamics can affect the perforated tile airflow distribution. Since only the plenum is modeled here, the results are strictly CALIBRATION AND EXPERIMENTAL VERIFICTION
applicable only to the case of uniform room pressure above the OF CFD MODEL
Floor Plan A with 0.23 m (9 in) and 0.61 m (24 in) deep As mentioned above, obstructions are not included in the plenums was used to assure that results were grid-independent. CFD models. While obstructions may significantly impact The maximum length of any side of any grid cell was airflow uniformity, the effects are typically fairly localized. systematically reduced until predicted tile airflow results Furthermore, given the fairly random nature of obstructions, it stabilized. Ultimately, a grid size was selected with the is difficult to quantify the effects of obstructions in any useful following characteristics: a maximum cell size of 15 cm (6 in), a minimum cell size of 2.5 cm (1 in), and a minimum of 8 cells in the plenum-depth direction. These grid settings were then Perforated tiles are modeled with the pressure-airflow applied to all cases. relationship given in Eq. (1) and the loss coefficients indicated Experimental Verification of CFD Model As discussed above, leakage airflow is driven by a pressure In order to verify the modeling methodology used in this difference and may vary widely across the data center floor. paper, the experimental results reported in [6] are used. One reasonable approach to modeling leakage would be to Specifically, one of the six layouts studied in [6] is used here represent the raised floor as a resistance with a particular loss for this verification and is depicted in Table 1 as Floor Plan B coefficient, which establishes the local leakage airflow based with the results shown in Fig. 3. The case studied is an array of on local pressure difference across the raised floor. While 4 x 15 perforated tiles situated between the 2 CRAC units. The physically realistic, this model is not ideally suited for present experiments were performed on a portion of a large raised-floor purposes because, following this approach, the total amount of data center located at IBM in Poughkeepsie, New York. This leakage airflow is not known a priori; it is an output from the area measured 6.06 (20 ft) x 20 m (66 ft) with a raised floor simulation. Therefore, the approach taken here is to simply height of 29.2 cm (11.5 in). The floor tiles were 610 mm (2 ft) remove (via a mass sink located just under the raised floor) the on a side with the position of the CRAC units as noted in the leakage airflow uniformly across the floor plan. This is plan view shown in the figure. The CRAC units were convenient for defining leakage rates (e.g. 0%, 20%, and 40%) positioned such that the momentum of the air exhausting from and is deemed adequate in light of the real variations in leakage the blowers within the CRAC units were directed toward each paths and other assumptions in the models. other, that is, the air was exhausting such that it collided in a region between the two CRAC units. To accurately measure CRAC's are modeled with fixed airflow rates. This is the flow exhausting from the tiles and to account for all the justifiable as the external (e.g., plenum and perforated tile) flow exiting the CRAC units, this testing area of the floor was resistance is much less than the internal CRAC resistance, blocked off around the perimeter of the testing area from the which may be 250 to 750 Pa (1 to 3 in H2O) or more. Blower raised floor down to the concrete subfloor. This 29.2 cm (11.5 outlets are modeled explicitly with the exception of Floor Plan I in) height (as measured from the subfloor to the bottom of the for which the airflow is specified uniformly over the entire tile) was carefully sealed with cardboard and duct tape. In CRAC footprint. Turning vanes are not included in any cases addition, any electrical or plumbing openings in this area were as the details of which vary from vendor to vendor. Furthermore, as turning vanes generally keep stronger jets in tact farther from the CRAC, they lead to decreased airflow The test results shown in Fig. 3 and Table 3 are with both uniformity compared to simply allowing the airflow from the CRAC units operating and neither CRAC unit having a turning CRAC to diffuse more evenly over the subfloor. vane. Each perforated tile is a nominally 25% open perforated tile. Actual measurements of the tile showed an array of 0.64 IT equipment which draw air directly from the plenum cm (0.25 in) diameter holes resulting in a 19.5 % opening, not ("plenum-ducted" in Table 1) are also modeled with fixed 25% as the manufacturer states. The impedance of the airflow rates and the (outward) flow is specified as leaving the perforated tiles were measured on a flow bench with the plenum uniformly over the entire IT equipment footprint. Note following results: that most IT equipment does not draw air directly from the plenum and therefore is not included in the plenum-only DP[Pa] = 419 {Q[m3/s]}1.99 (7a) Copyright 2005 by ASME reverse direction impinges on the left wall (at x = 0), turns 180 deg., and exits from the tiles in the middle. Most of the fluid exiting the unit B is discharged as a jet towards unit A. A small DP[in H20] = 4.05x10-7 {Q[cfm]}1.99 (7b) amount of fluid impinges on the east wall (at x = 20 m), turns around, and is also exhausted through the tiles in the middle. The longitudinal velocities (directed along the x axis) are larger near unit B, causing large pressure variation in this region. The peak in the airflow velocity distribution is located closer to unit A and corresponds to the location where the two opposing CFD SIMULATION RESULTS AND CONCLUSIONS
Airflow (cfm) -50 Example Showing Airflow Uniformity is Independent of Table 4 summarizes the scenarios considered. For Floor Plans A and H, corresponding scenarios were modeled with the total airflow rate halved and doubled. CRAC airflow, leakage airflow, and ducted-IT airflow were all halved or doubled (though the percentage of leakage airflow is constant) although only mean airflow per tile is shown in the Table 4. Airflow (cfm) -50 Fig re 3 – CFD Airflow uniformity is, indeed, largely independent of flow rate. The small variations that do exist reflect the fact that all the losses in the plenum do not scale identically with airflow The air momentum is such that the air streams from both CRAC units collide near the center of the perforated tile region. Effect of Parameters Considered on Uniformity The flow distribution shows a fairly symmetric distribution as would be expected with the minimums occurring nearest the Table 5 shows all results from the main matrix of 240 CRAC units and the maximum near the center of the perforated scenarios considered. The top half of the table covers 25%- tile region. Although some asymmetry is evident, it may be open tile scenarios; the bottom half covers 56%-open tile due to the differences in total airflow rates from the CRAC scenarios. Results show considerable tile-to-tile airflow units, from the asymmetric location of CRAC blower outlets, or variations. In fact, all the floor plans exhibit some backflow the asymmetry of non-perforated tiles adjacent to the CRAC under at least one scenario with 56%-open tiles. Selected units. (As many of these details were not included in the CFD results are discussed in greater detail below. model, the only sources of asymmetry in the CFD model are the asymmetry of the room layout and asymmetric location of CRAC blower outlets.) The flow from some of the perforated tiles nearest the CRAC units showed some flow downward into the raised floor plenum. The predicted flow rates from the model are in good agreement with the measured values. Although not shown, the flow exiting CRAC unit A (see Table 1) splits into two streams: one moving in the forward direction (toward the right) and the other in the reverse direction. The fluid in the forward stream exits from the tiles close to unit A. The stream flowing in the Copyright 2005 by ASME Figure 4 shows the effect of floor plan on airflow uniformity with 20% leakage and a 0.61 m (24 in) plenum. Other than Floor Plan B (which was primarily intended for validating the CFD model, not as a practical layout) the standard deviation from the mean is on the order of 10% for 25%-open tiles and 50% for 56%-open tiles. The probability distributions for these values of standard deviation are also shown in Fig. 4. A reasonable spread around the mean is associated with the 25%-open tiles while the spread associated with the 56%-open tiles is very broad. 68% of the 25%-open tiles will be within ±10% of the mean airflow per tile whereas 68% of the 56%-open tiles are within only ±50% of the mean. So, from a standard deviation perspective, most 25%-open tile floor plans deliver reasonable uniformity while most 56%-open tile floor plans do not. Even with 25%-open tiles and small standard deviations, some floor plans, C and F in particular, have considerable minimum and maximum variations with some backflow occurring under most scenarios. From Table 5 and Fig. 4, we see that the floor plans achieving the best uniformity are those in which the CRAC's supply air in the direction parallel to the perforated-tile rows with room and CRAC layouts, which are rectangular and symmetric. Floor Plan G performs very well, however, the large fraction of plenum-ducted IT equipment makes this floor plan fairly atypical. The best typical layout is Floor Plan A, which is a standard 7-tile pitch hot aisle/cold aisle layout [1]. Standard Deviation Distribution Density u e 4 – Airflow U g and 0.61 m (24 in) P Figure 5 shows the effect of plenum depth with 25%-open tiles and 20% leakage. The five floor plans shown are fairly representative of the more typical (of real data center) floor plans considered. Uniformity improves only modestly with plenum depth except for Floor Plan C (in which the CRAC's supply air in the direction perpendicular to the perforated tile rows). Considering the results of all floor plans studied (Table 5), there is no great benefit to increasing the plenum depth beyond 0.61 m (24 in). As plenum obstructions were not included in the models, actual plenums should be designed with clear airflow space of 0.61 m (24 in) or more. Copyright 2005 by ASME with backflow first occurring at about 45 tile widths. Although not shown in Fig. 7, the maximum tile airflow always occurs near the center of the room with the minimum tile airflow occurring at the perforated tiles located nearest the CRAC's. There is no maximum distance across which air can be supplied; however, beyond 20-30 tile widths uniformity becomes a limiting factor. As more and more tiles are added, the mean airflow per tile and the average plenum pressure drop. Standard Deviation 10% However, the structure of the supply airflow (jets) near the CRAC's does not change substantially as perforated tiles are added. Consequently, pressure disturbances near the CRAC become larger relative to tile resistance (which scale with the airflow rate squared) as the row is lengthened. Although somewhat counterintuitive, supplying additional airflow will not help, as the uniformity is independent of airflow rate. Finally, it is stressed that these comments are made with respect to Floor Plan A of Table 1. If, instead, CRAC's were located only on one side of the room (with a wall at the other) we could expect uniformity to become poor beyond 10-15 tile widths. Figure 6 shows the effect of leakage airflow with 25%- open tiles and 0.61 m (24 in) plenums. The same five floor plans as discussed above in the context of plenum depth are again taken as representative. Uniformity degrades with increasing airflow leakage; the net effect is similar in magnitude to the effect of plenum depth over the respective parameter ranges considered here. Recall that the greater the tile resistance relative to airflow path resistance, the better the uniformity. Airflow lost to leakage in the plenum effectively increases the resistance (creates a greater pressure loss) along the airflow path making the perforated tile resistance less dominant. While minimizing leakage airflow will improve uniformity, from a cooling perspective the airflow is not completely wasted as it goes to cooling the room and equipment in the vicinity of the leakage openings. ig e 7 – Airflow U DESIGN RECOMMENDATIONS TO ACHIEVE
Based on the above findings, the following design recommendations are made with respect to maximizing perforated tile airflow uniformity: • Use only the more-restrictive perforated tiles (e.g. 25% open) for general deployment. High-flow perforated tiles (e.g. 56% open) should be used only in special circumstances. Examples include: the number of high- flow tiles constitute only a small fraction of perforated tiles present, an isolated plenum has been created and uniformity verified through measurement, and hot IT- equipment exhaust is ducted directly back to the CRAC's (so that uniform air delivery is not critical). The effect of row length on airflow uniformity was • Create standard hot aisle/cold aisle layouts (like Table 1, investigated using Floor Plan A with a 0.61 m (24 in) plenum Floor Plan A) utilizing rectangular, symmetric layouts. and 20% leakage. The room was made wider and narrower by Locate CRAC's at the end of the hot aisle (far from adding and removing perforated tiles while keeping all other perforated tiles) and avoid rows longer than about 20 geometry and conditions fixed. Figure 7 shows the results. perforated tiles or about 10 perforated tiles for layouts with Uniformity degrades fairly quickly beyond 20-30 tile widths CRAC's at only one end with a wall at the other. Copyright 2005 by ASME • Design plenums for clear airflow space of 0.61 m (24 in) or • Minimize leakage airflow through non-perforated tile openings in the raised floor. • Keep chilled water pipes and cables away from the exhaust regions of A/C units and only use turning vanes where the A/C units are used in an inline layout where each is used to boost the static pressure of the next in line [6]. • Do not simply increase airflow rate without addressing the other factors listed above, as this will not improve REFERENCES
"Thermal Guidelines for Data Processing Environments", American Society of Heating, Refrigerating, and Air-Conditioning Engineers (ASHRAE) 2004 [2] "Optimizing Data Centers for High-Density Computing", Hewlett-Packard Technology Brief, The Uptime Institute. "Alternating Cold and Hot Aisles Provides More Reliable Cooling for Server Farms". [4] VanGilder, J. and Lee, T., 2003, "A Hybrid Flow Network-CFD Method for Achieving Any Desired Flow Partitioning Through Tiles of a Raised-Floor Data Center", InterPACK 2003, Maui, Hawaii. [5] Kang, S., Schmidt, R., Kelkar, K., Patankar, S., "A Methodology for the Design of Perforated Tiles in Raised Floor Data Centers Using Computational Flow Analysis", IEEE-CPMT Journal, Vol. 24, No. 2, pp. 177- [6] Schmidt, R. et al, 2001, "Measurements and Predictions of the Flow Distribution Through Perforated Floor Tiles In a Raised-Floor Data Center", InterPACK 2001, Kauai, [7] Holman, J.P., 1989, Experimental Methods for Engineers, McGraw-Hill, Inc., New York., pp.49-50, 57. FLOVENT V5.1 Software by Flomerics, Flomerics Ltd., 81 Bridge Road, Hampton Court, Surrey KT8 9HH, UK, Copyright 2005 by ASME Airflow Uniformity through Perforated Tiles in a Raised-Floor Data Center Click on icon to link to resource For feedback and comments about the content of this white paper: Data Center Science Center [email protected] If you are a customer and have questions specific to your data center project: Contact your Schneider Electric representative Schneider Electric – Data Center Science Center White Paper 121 Rev 1 10

  • Word Bookmarks


    Órgano de difusión interno de la AMPF y AMPC Diciembre de 2010 Publicación de distribución gratuita para el asociado Familia - País - Sociedad Sede Central Tte. Gral. Juan D. Perón 1379/83 - CABA - (011) 4124-9910 y rot. - - [email protected] El Correo Solidario Comisión Directiva Sumario Staff Imagen de tapa: Danza de las Cintas (Detalle) Manuel Arce


    Journal of Gastroenterology and Hepatology Research Journal of GHR 2013 March 21 2(3): 449-457 ISSN 2224-3992 (print) ISSN 2224-6509 (online) Interaction of Insulin with Prokinetic Drugs in STZ-induced Diabetic Mice Mohamed A Fouad, Hekma A Abd El latif, Mostafa E El Sayed Mohamed A Fouad Shalaby, Department of Pharmacology, R&D

    Copyright © 2008-2016 No Medical Care