2012 - Rack Level Modeling of Air Flow Through Perforated Tile in a Data Center

download 2012 - Rack Level Modeling of Air Flow Through Perforated Tile in a Data Center

of 11

Transcript of 2012 - Rack Level Modeling of Air Flow Through Perforated Tile in a Data Center

  • 8/13/2019 2012 - Rack Level Modeling of Air Flow Through Perforated Tile in a Data Center

    1/11

    1 Copyright 2012 by ASME

    Proceedings of the ASME 2012 International Mechanical Engineering Congress & ExpositionIMECE2012

    November 9-15, 2012, Houston, Texas, USA

    IMECE2012-85203

    RACK LEVEL MODELING OF AIR FLOW THROUGH PERFORATED TILE IN A DATA

    CENTER

    Vaibhav K. ArghodeG.W. Woodruff School of Mechanical Engineering

    Georgia Institute of TechnologyAtlanta, Georgia, USA

    Pramod KumarDepartment of Mechanical Engineering

    Indian Institute of ScienceBangalore, Karnataka, India

    Yogendra JoshiG.W. Woodruff School of Mechanical Engineering

    Georgia Institute of TechnologyAtlanta, Georgia, USA

    Thomas S. WeissTriad Floors Inc.

    Denver, Colorado, USA

    Gary MeyerTriad Floors Inc.

    Denver, Colorado, USA

    ABSTRACTEffective air flow distribution through perforated

    tiles is required to efficiently cool servers in a raised

    floor data center. We present detailed computational

    fluid dynamics (CFD) modeling of air flow through a

    perforated tile and its entrance to the adjacent server

    rack. The realistic geometrical details of the perforated

    tile, as well as of the rack are included in the model.

    Generally models for air flow through perforated tiles

    specify a step pressure loss across the tile surface, or

    porous jump model based on the tile porosity. Animprovement to this includes a momentum source

    specification above the tile to simulate the acceleration

    of the air flow through the pores, or body force model.In both of these models geometrical details of tile such

    as pore locations and shapes are not included. More

    details increase the grid size as well as the

    computational time. However, the grid refinement can

    be controlled to achieve balance between the accuracy

    and computational time. We compared the results from

    CFD using geometrical resolution with the porous

    jump and body force model solution as well as with

    the measured flow field using Particle Image

    Velocimetry (PIV) experiments. We observe thatincluding tile geometrical details gives better results as

    compared to elimination of tile geometrical details and

    specifying physical models across and above the tile

    surface. A modification to the body force model is also

    suggested and improved results were achieved.

    Key words: High density rack; Perforated tile;

    Air flow distribution; Geometrical resolution; Porous

    jump model;Body force model

    INTRODUCTIONIn recent past, many investigations have been

    reported on modeling of data center air flows [2-11].

    Some of the investigations only focused on the under

    floor region (plenum) modeling [3-5] and they

    imposed a step pressure loss across the tile surface, or

    porous jump model thereby eliminating any

    geometrical details of the tile from the model. The

    pressure loss factor (K=P/0.5Vin2) was calculated

    based on the tile porosity (open area ratio) [1]. Thefocus of these works was to predict the air flow rate

    emerging from the plenum. The suggested models

    include a 0D model where a uniformly pressurized

    plenum was considered [3] or 1D model where the

    effects of under floor pressure variation as well as

    frictional resistance in plenum was considered. These

    simplified model results in manageable grid size and

    reduced computational time for modeling large data

    centers. Such investigations are helpful for

    understanding the overall air supply pattern in the data

    center with objective of supplying the air flow rate as

    per the requirement of the adjacent rack. It was also

    suggested that the air flow design problem is based onthe under floor flow features and supplying the

    required amount of air is sufficient to effective cool

    the adjacent server rack [2]. However experimental

    investigation of air flow emerging from perforated tile

    and entering the rack suggests that even though the

    supplied air flow rate matches the rack requirement,

    issues such as air bypass and room air entrainment

    may be present [13-14]. Hence modeling of above

  • 8/13/2019 2012 - Rack Level Modeling of Air Flow Through Perforated Tile in a Data Center

    2/11

    2 Copyright 2012 by ASME

    floor system is deemed useful and has been considered

    in investigation such as [8-11].

    In a more comprehensive investigation, both

    under floor and above floor regions have been

    investigated together [6]. In this investigation the two

    regions were connected by perforated tiles where a

    step pressure loss is specified through the tile surface

    (porous jump model). A review of the computational

    investigations pertaining to data center air flows could

    be found in [7].

    An improvement in the tile flow model was

    suggested in [10-11]. However, in these investigations

    the focus was limited to the above floor surface only

    where designated air flow rates were specified at the

    tile inlet. It was mentioned that in the porous jump

    model (which considers uniform velocity based on

    total open area of tile), the momentum rise of air due

    to pressure loss across the tile surface is not captured

    and hence in these investigations additional body force

    (momentum source) was specified above the tile

    surface to simulate higher momentum of the air flowabove the tile surface (body force model). This model

    showed improved results as compared to the porous

    jump model [10-11]. Other strategies such as modeling

    the tile with an equivalent reduced single open area to

    simulate higher momentum or multiple openings were

    also suggested [10]. Multiple openings with some tile

    geometrical details (much larger pore length scales as

    compared to the actual pore size) was also investigated

    and fair agreement with measurements was obtained

    [10]. In this case higher air entrainment and faster

    mixing was observed as compared to the body force

    model which eliminated pore geometrical details.

    This suggests that further inclusion of finergeometrical details can be helpful to closely capture

    the flow pattern above the floor and at rack entry.

    Hence, in the present work we report the effect of

    including more resolved tile geometrical details on the

    solution accuracy at the expense of increased

    computational effort. Note that a balance between the

    solution accuracy and computational effort needs to be

    considered while investigating data center air flows.

    We also present a modification to the body force

    model which could be useful to simulate the flow field

    with much reduced computational efforts and it also

    shows promise for further improvements to develop

    simplified models which will reduce thecomputational time while still maintaining acceptable

    accuracy.

    COMPUTATIONAL SETUPFigure 1 shows the server simulators, rack and the

    perforated tile used in the experimental investigation

    [13-14]. The 42U (1U = 4.45 cm) rack houses four

    server simulators each having height of 10U (see

    Figure 1(a)). There is a gap of about 2.54 cm (1 inch)

    between the bottom of the rack and the solid tile

    surface below the rack (see Figure 1(b)). Server

    simulators have a front plastic grill with porosity of

    about 37% [12]. Each server simulator has four fans

    mounted on a plate. The fan-plate is located at a

    distance of 25.4 cm (10 inch) from the front of server

    simulator. The fans are centered and spaced 5.08 cm (2

    inch) apart from fan outer edge. Each fan has an outer

    diameter of 16.83 cm (6.625 inch) and 8.26 cm (3.25

    inch) hub diameter. Each server simulator has fan

    speed dial settings to set desired flow rates through the

    rack. Further details of the server simulator could be

    found in [12].

    The tile used for the investigation has dampers on

    the bottom and a perforated plate on the tile top

    surface. The dampers have multiple rectangular

    openings (see Figure 1(d)) and the tile top has

    openings of different shapes with length scales of

    about 1.27 cm (0.5 inch) (see Figure 1(c)). The

    distance between the dampers and the tile top is about

    3.81 cm (1.5 inch).Figure 2 shows the computational domain and

    modeled geometrical tile details. The computational

    domain considers only one tile and one rack as shown

    in the Figure 2(a). All the geometrical features are

    modeled as orthogonal shapes. Circular fans with hub

    are modeled as square openings with square hub. The

    rack geometry after the fan plate is not included in the

    model and the fans are considered as exhaust fans with

    target mass flow boundary condition. In this case the

    pressure at the fan surface is adjusted so as to reach a

    pre-assigned mass flow rate through the fan surface.

    The swirl generated by the fans is not included in the

    model.Figure 3 shows the dimensional details of the

    computational domain. The grill at the rack inlet is

    modeled as thin rectangular openings of about 1.27 cm

    (0.5 inch) height. The impermeable strips adjacent to

    the openings have height of 2.54 cm (1 inch) hence

    giving an overall porosity of 33% (see Figure 3(d)).

    All four server simulators are separated by

    impermeable walls as shown in Figure 3(a). The gap

    of 2.54 cm (1 inch) between the bottom of the rack and

    solid tile below it is also considered in the model and

    pressure inlet boundary condition is specified at the

    inlet of the gap (see Figure 3(a)).

    Plenum is included in the model by specifyingmass flow rate boundary condition below the tile

    surface with symmetry boundary condition on the

    plenum walls (see Figure 2(a), 3(a), 3(b)). For the

    present investigation the plenum is fairly deep (~ 0.9

    m or 3 ft) and hence the effect of directional flow

    under the tile surface is expected to be less in the

    present case [2]. Moreover, two down flow CRAC

    units were used which were located on either sides of

    the cold aisle (along the length). This would further

  • 8/13/2019 2012 - Rack Level Modeling of Air Flow Through Perforated Tile in a Data Center

    3/11

    3 Copyright 2012 by ASME

    suppress the directional flow characteristics in the

    plenum, see [13-14] for data center layout. Hence,

    mass flow condition is specified below the tile surface

    and not on the sides of the modeled plenum. The aisle

    top is specified as pressure outlet and the aisle

    center is specified as pressure inlet. The aisle

    center is the aisle surface facing the rack inlet. The

    sides of the aisle are specified as pressure inlet with

    some resistance to the air flow (pressure loss

    coefficient, K = 10). This condition was used due to

    the anticipated resistance to the flow entering the

    computational domain because of the presence of

    adjacent tiles. Refer [13-14] for further details on tile

    layout for the experimental investigation. The value of

    K was appropriately adjusted to simulate the flow field

    features obtained by experiments. However, further

    investigation is required to arrive at an appropriate

    value for K based on the physical set-up and operating

    conditions.

    The modeled tile geometry is shown in Figure

    2(b-f). The tile top surface is modeled as 900 squarepores of 1.27 cm (0.5 inch) side. Note that the tile top

    surface has pore shapes which have curved edges,

    however in our investigation we observe that the

    curved pores could be modeled as orthogonal shapes

    with same hydraulic diameter [15]. The spacing

    between two square pores on the tile top is about 0.64

    cm (0.25 inch). The porosity of the modeled tile top

    surface is about 39%. The damper openings are

    modeled as 36 rectangular openings, each with size of

    3.18 cm (1.25 inch) 13.34 cm (5.25 inch). The

    distance between two damper openings is 0.64 cm

    (0.25 inch) along the longitudinal direction and 3.18

    cm (1.25 inch) in the transverse direction. The porosityof damper openings is 41%. The tile top and the

    dampers are spaced 3.81 cm (1.5 inch) apart (see

    Figure 3(a)).

    The computational domain is meshed with full

    rectilinear grid. The region near the perforated tile is

    appropriately refined such that each square pore on the

    tile top has 8 8 cells. In this case grid refinement was

    performed using hanging-node approach. Simulations

    suggest that having 8 8 cells in each pore gives

    reasonably close result with a grid independent

    solution [15]. Note that appropriate balance between

    grid refinement and solution accuracy is sought here.

    For the case using geometrical details the grid size isabout 4.2 million cells. Without grid refinement, for

    the cases using simplified models such as porous jump

    or body force models the grid size is about 1 million

    cells.

    The isothermal flow field was solved using a

    steady state, finite volume based method and SIMPLE

    algorithm was used for pressure velocity coupling

    [16]. Symmetry with only one-half the geometry was

    used to reduce the computational time. k- realizable

    turbulence model was used with inlet turbulence

    intensity of 5%. For convective term discretization,

    second order upwind scheme was used. Convergence

    was assumed when the normalized residuals for all the

    variables were less than 1E-04 and the selected

    velocity monitors have stabilized within variation of

    1%. Commercial CFD software FLUENT was used for

    the simulations.

    The results are compared with the experiments

    [13-14], and with the simplified models specifying

    porous jump boundary condition as well as body force

    source above the tile surface.

    Porous jump model:

    Porous jump model specifies step pressure loss

    across the tile surface [2-9]. Generally, the magnitude

    of pressure loss factor is calculated based on the

    porosity of the tile. The pressure loss specification has

    a form similar to that given below in Equation (1-2)

    [1].

    = (1) = +1 (2)

    = , = , = = It may be noted that use of this model results in no

    alteration in the velocity field across the tile surface.

    However, there is acceleration of air though the pores

    resulting in increased momentum flux, attempted to be

    captured by the body force model. It may be noted that

    the tile has two surfaces in series in the path of airflow. The bottom surface has tile dampers and top

    surface has the perforations as shown in Figure 2(b-f).

    The pressure drop across theses surfaces, as obtained

    by correlation (equation 1-2), is added to arrive at the

    final pressure drop. Note that in actual case the flow is

    not uniform after passing through the dampers and this

    may affect the overall pressure drop from the tile. For

    the present case the equivalent K across the two tile

    surfaces is 16.5. The equivalent porosity of the two

    layers is calculated as 31%. The value of K across the

    rack inlet (front grill of server simulators) is 14.3

    corresponding to the grill porosity of 33%.

    Body force model:

    In this model a momentum source is specified

    above the tile surface to simulate the acceleration of

    air passing through the pores in the tile (see equation

    (3-5)) [10-11]. The magnitude of the momentum

    source was calculated based on the difference of

    momentum through the tile pores as compared to the

    momentum of air flow in case of uniform velocity as

    in the case in porous jump model (see Equation (3)).

  • 8/13/2019 2012 - Rack Level Modeling of Air Flow Through Perforated Tile in a Data Center

    4/11

    4 Copyright 2012 by ASME

    The momentum of air flow through the pores was

    calculated assuming top-hat velocity profile through

    the pores (see Equation (4)).

    = (3) =

    (4)

    = (5), "" = =

    Modified Body force model:

    The body force model suggested previously

    considers a top hat velocity profile at the tile pores

    while calculating the excess momentum of air above

    the tile surface. However, even higher momentum of

    the jets is expected as in reality there will be a velocity

    profile present at the tile pores instead of a top-hat

    profile. Hence, we suggest a modification to the bodyforce model. In the modified body force model the

    momentum rise across the tile surface is calculated

    based on the pressure loss as obtained from the

    correlations (see Equation (6)).

    = = = + 1 (6), Once the momentum averaged velocity (Vpore)

    through the tile surface is calculated the momentum

    source term can be calculated as suggested by the

    body force model (see Equation (5)). Note that the

    modified body force model will have higher

    magnitude of momentum source as compared to the

    body force model.

    RESULTS AND DISCUSSIONTable 1 summarizes the cases studied. The air

    flow rate across the rack corresponds to a high power

    density rack with 15 kW of heat dissipation and bulk

    air temperature rise of 10 C across the rack. The tile

    flow rate is varied from 0% upto 100% of the rack air

    flow rate and the values are listed in Table 1. Further

    details of the experimental conditions could be found

    in [13-14].

    Table 1. Different cases investigated.

    Case

    #

    Rack flow

    rate (m3/s,

    CFM)

    Tile flow

    rate (m3/s,

    CFM)

    Tile flow /

    Rack flow

    1 1.224, 2594 0, 0 0%

    2 1.224, 2594 0.234, 496 20%

    3 1.224, 2594 0.754, 1598 60%

    4 1.224, 2594 1.224, 2594 100%

    Figure 4 shows the pressure loss obtained across

    the tile surface from CFD simulations. In this case the

    tile geometrical details are included in the model as

    shown in Figure 2(b-f). The correlation used forcomparison is given in Equation (2) corresponding to

    equivalent tile porosity of 31%. Figure 4 suggests a

    good comparison between the correlation and CFD

    simulations. However, note that we assume addition of

    pressure loss across the two tile surfaces based on the

    correlations and in actual case this may not represent

    the actual pressure loss. Value of K across the tile for

    cases 2, 3 and 4 are also very close to each other

    suggesting that the pressure loss factor does not vary

    significantly with respect to flow rates investigated

    here. Note that case 1 is not included, as there is no

    flow through the tile in this case.

    Figure 5-8 shows the comparison between

    experimental (PIV) and computational (CFD) results

    for cases 1-4 (see Table 1). Pathline plots are used for

    visualization of the flow field. In the CFD simulations,

    massless particles released from the lines at

    boundaries (tile, aisle center and aisle top, see

    Figure 5-8) and are tracked and plotted. The plots

    correspond to the symmetry plane positioned along the

    height of the rack. Note that in the CFD model

    considering geometrical resolution, the modeled

    geometry of the grill at the rack inlet is included. In

    the porous jump model pressure loss is specified

    across the grill surface.

    Figure 5 shows the results for case 1 (see Table 1).

    In this case there is no flow through the tile. From thefigure we observe that the flow field measured by PIV

    and calculated using CFD are similar and entrainment

    of air from aisle top as well as the aisle center is

    present in both the experimental and computational

    results. The top region of the rack inlet shows air

    entrainment from the aisle top (see figure 5(a)). Figure

    5(b-e) suggests that the resolution of grill geometry

    (rack inlet) does not significantly influence the

    computational solution, as compared to using the

    porous jump model across the grill for the present

    case. This also suggests that it may be beneficial to use

    porous jump boundary condition across the rack inlet

    to save computational effort for this case.Figure 6 shows the flow field for case 2 (see Table

    1). In this case the flow rate through the tile is about

    20% of the air flow rate through the rack. Due to

    insufficient supply of air from the perforated tile, the

    rack also entrains the air from aisle top as well as aisle

    center as shown in Figure 6(a). The air supplied from

    the tile reaches only the lower portion of the rack,

    while near the top the rack entrains air from aisle top.

    The middle portion of the rack mostly entrains air

  • 8/13/2019 2012 - Rack Level Modeling of Air Flow Through Perforated Tile in a Data Center

    5/11

    5 Copyright 2012 by ASME

    from the aisle center. These flow field features are also

    captured by all the computational models (see Figure

    6(b-e)). The body force model and modified body

    force model show that the tile air flow reaches higher

    along the rack height, as compared to the porous jump

    model (see Figure 6(b) and 6(c)). Inclusion of body

    force source above the tile surface increases the

    momentum of bulk air in the region above the tile

    surface, and hence results in higher reach of air to the

    rack inlet. The momentum source region is between

    the yellow line and tile surface (see Figure 6(c-d)).

    The result obtained with geometrical resolution is very

    similar to the result obtained by the body force and

    modified body force models.

    The flow field for case 3 (60% of rack air flow

    through tile, see Table 1) is shown in Figure 7. From

    Figure 7(a) we note that even though the air flow rate

    through the tile is lower than the rack air requirement,

    a portion of the supplied air escapes from the aisle top.

    We can also observe delayed entry of the tile air at the

    lower portion of rack inlet (see bottom right corner ofFigure 7(a)). This was attributed to the higher

    momentum of air emerging from the tile [13-14]. Air

    entrainment from the aisle center could also be noticed

    for this case (see bottom left corner of Figure 7(a)).

    The porous jump model is not able to capture these

    phenomena and the air flow from the tile reaches only

    up to a certain height of rack (see Figure 7(b)). Porous

    jump model also suggests entrainment of air from aisle

    top which is not observed experimentally. The porous

    jump model is also not able to capture the air

    entrainment from aisle center near the tile surface

    (bottom left corner) and delayed entry of air in the

    lower portion of rack inlet (bottom right corner).When momentum source is included above the

    tile surface (body force model) the tile air flow reaches

    higher as compared to the porous jump model (see

    Figure 7(c)). However air entrainment from the aisle

    top could be observed for the body force model too,

    which is not present in the experimental results. Body

    force model shows minor improvement in the

    computed flow field. In this case, small entrainment of

    air from the aisle center near tile (bottom left corner)

    and slight delay of entry of air in lower portion of rack

    could be observed (bottom right corner).

    The modified body force model shows further

    improvement in the results possibly due to highermagnitude of momentum source specification above

    the tile surface (see Figure 7(d)) as compared to the

    body force model. It may be noted that even with the

    modified body force model, the flow field is

    significantly different from the experimental results.

    The flow field with the model that included the

    geometrical details of the tile is shown in Figure 7(e).

    From the figure we observe that the flow field features

    are closer to the experimental results. A portion of air

    flow is observed to escape from the aisle top. This

    model also captures the delayed entry of air to the

    lower portion of the rack inlet. The air entrainment

    from aisle center could also be observed from Figure

    7(e).

    Figure 8 shows the flow field features for case 4

    (see Table 1). In this case, the air flow rate through the

    tile is equal to the air flow rate through the rack. Due

    to higher momentum of air jet, significant portion of

    the tile air is observed to bypass the rack through the

    aisle top (see Figure 8(a), refer to [14] for the

    discussion). The air entrainment through the aisle

    center and delayed entry of tile air to the rack could be

    prominently observed. The porous jump model (see

    Figure 8(b)) shows the tile air reaching near the aisle

    top, however, the prominent flow field features

    measured experimentally, could not be captured by

    this model. The body force model, as well as modified

    body force model while showing improved prediction,

    could not compare closely with the experimental

    results (see Figure 8(c-d).Figure 8(e) shows the flow field obtained by

    resolving the tile geometry, which is closer to the

    experimental results. This model is able to capture

    significant bypass of the tile air through the aisle top,

    as well as entrainment of air though aisle center near

    the tile (bottom left corner). The delayed entry of the

    air near the bottom portion of the rack could be clearly

    seen from bottom right corner of Figure 8(e). This

    suggests that simplified geometrical resolution shows

    promise for better solution accuracy and further

    improvements in the body force, or development of

    other simplified models is required to capture the flow

    field closely.

    CONCLUSIONSExperimentally measured flow field in data center

    cold aisles were compared with various computational

    modeling approaches. The simplified flow models

    through floor tiles, such as porous jump and body

    force models suggests fair comparison with the

    experimental results at lower tile air flow rates.

    However, at higher air flow rates, required for high

    power density racks, these simplified models do not

    capture the flow field features accurately. A modified

    body force model is suggested, which calculates the

    momentum rise based on the prescribed pressure lossacross the tile surface. While this model shows

    improvement in the results, it still does not compare

    well with the experimental results at higher tile air

    flow rates. Including the geometrical details of the tile

    provides better comparison with the experimental

    results. This model is able to capture the prominent

    flow features, such as bypass of tile air though the

    aisle top, as well as entrainment of air from aisle

    center and delayed entry of air to the rack inlet. The

  • 8/13/2019 2012 - Rack Level Modeling of Air Flow Through Perforated Tile in a Data Center

    6/11

    6 Copyright 2012 by ASME

    present investigation suggests that a fairly coarse grid

    could be used, with more geometrical details to

    achieve better solution accuracy as compared to

    specifying simplified models across the tile surface.

    Improvements in the simplified models show scope for

    future exploration to achieve a better balance between

    the desired solution accuracy and reduced

    computational time.

    ACKNOWLEDGMENTSThis research is supported by Triad Floors, Inc. and

    the Consortium for Energy Efficient Thermal

    Management (CEETHERM). The support is gratefully

    acknowledged.

    REFERENCES[1] Freid, E., Idelchik, I. E., Flow Resistance, A

    Design Guide for Engineers, Hemisphere, New York,

    1989.

    [2] Patankar, S. V., Airflow and Cooling in a Data

    Center, Journal of Heat Transfer, 2010, Vol. 132, pp.073001-1-17.

    [3] Kang, S., Schmidt, R., Kelkar, K. M.,

    Radmehr, A., Patankar, S. V., A Methodology for the

    Design of Perforated Tiles in Raised Floor data

    Centers using Computational Flow Analysis, IEEE

    Transactions on Components and Packaging

    Technologies, 2001, Vol. 24, pp. 177-183.

    [4] Karki, K. C., Patankar, S. V., Airflow

    Distribution Through Perforated Tiles in Raised-Floor

    Data Centers, Building and Environment, 2006, Vol.

    41, pp. 734-744.

    [5] Schmidt, R. R., Karki, K. C., Kelkar, K. M.,

    Radmehr, A., Patankar, S. V., Measurements andPredictions of the Flow Distribution through

    Perforated Tiles in Raised-Floor Data Centers,

    International Electronic Packaging Technical

    Conference and Exhibition, July 8-13, 2001, Kauai,

    Hawaii, USA.

    [6] Rambo, J., Joshi, Y., Convective Transport

    Process in Data Centers, Numerical Heat Transfer

    Part A, 2006, Vol. 49, pp. 923-945.

    [7] Rambo, J., Joshi, Y., Modeling of Data Center

    Airflow and Heat Transfer: State of the Art and Future

    Trends, Distributed Parallel Databases, 2007, Vol. 21,

    pp. 193-225.

    [8] Cruz, E., Joshi, Y., Iyengar, M., Schmidt, R.,Comparison of Numerical Modeling to Experimental

    Data in a Small Data center Test Cell, International

    Electronic Packaging Technical Conference and

    Exhibition, July 19-23, 2009, San Francisco,

    California, USA.

    [9] Iyengar, M., Schmidt, R. R., Hamann, H.,

    VanGilder, J., Comparision between Numerical and

    Experimental Temperature Distributions in a Small

    Data Center Test Cell, International Electronic

    Packaging Technical Conference and Exhibition, July

    8-12, 2007, Vancouver, British Columbia, Canada.

    [10] Abdelmaksoud, W. A., Khalifa, H. E., Dang,

    T. Q., Elhadidi, B., Schmidt, R. R., Iyengar, M.,

    Experimental and Computational Study of Perforated

    Floor Tile in Data Centers, Intersociety Conference on

    Thermal Phenomena, Jun 2-5, 2010, Las Vegas, USA.

    [11] Abdelmaksoud, W. A., Khalifa, H. E., Dang,

    T. Q., Schmidt, R. R., Iyengar, M., Improved CFD

    Modeling of a Small Data Center Test Cell,

    Intersociety Conference on Thermal Phenomena, Jun

    2-5, 2010, Las Vegas, USA.

    [12] Nelson, G., Development of an

    Experimentally-Validated Compact Model of a Server

    Rack. MS Thesis - 2007, Georgia Institute of

    Technology, GA, USA.

    [13] Kumar, P., Joshi, Y., Experimental

    Investigations on the Effect of Perforated Tile Air Jet

    Velocity on Server Air Distribution in a High Density

    Data Center, Intersociety Conference on Thermal

    Phenomena), Jun 2-5, 2010, Las Vegas, USA.[14] Kumar, P., Sundaralingam, V., Joshi, Y.,

    Dynamics of Cold Aisle Air Distribution in a Raised

    Floor Data Center, Thermal Issues in Emerging

    Technologies, Dec 19-22, 2010, Cairo, Egypt.

    [15] Arghode, V. K., Joshi, Y., Modeling

    Strategies for Air flow Through Perforated Tile in a

    Data Center, IEEE Transactions on Components,

    Packaging and Manufacturing Technology, In

    Preparation 2012

    [16] Patankar, S. V., Numerical Heat Transfer and

    Fluid Flow, Hemisphere, New York, 1980.

  • 8/13/2019 2012 - Rack Level Modeling of Air Flow Through Perforated Tile in a Data Center

    7/11

    7 Copyright 2012 by ASME

    (c) Tile top

    (a) Server simulators (b) Rack (d) Dampers

    Figure 1. (a-b) Rack and (c-d) tile details under investigation.

  • 8/13/2019 2012 - Rack Level Modeling of Air Flow Through Perforated Tile in a Data Center

    8/11

    8 Copyright 2012 by ASME

    (b) Tile top (c) Mesh for each square

    pore in tile top

    (d) Dampers (e) Mesh for eachopening in dampers

    (a) Computational domain (f) Perforated tile with dampers

    Figure 2. (a) Computational domain and (b-f) the modeled tile.

  • 8/13/2019 2012 - Rack Level Modeling of Air Flow Through Perforated Tile in a Data Center

    9/11

    9 Copyright 2012 by ASME

    Figure 3. Details of the computational domain and set-up.

    Figure 4. Pressure drop across the tile obtained by CFD with tile geometrical resolution.

    . . . .

    (.

    )

    (a) Side view (b) Front view (c) Fans (rack exit) (d) Grill (rack inlet)

    .

    ()

    .

    .

  • 8/13/2019 2012 - Rack Level Modeling of Air Flow Through Perforated Tile in a Data Center

    10/11

    10 Copyright 2012 by ASME

    Velocity

    (m/s)

    (a) PIV (b) Porous jump (c) Body force (d) Modified

    body force

    (e) Geometrical

    resolution

    Figure 5. Comparison of PIV and CFD results for case 1 (rack flow = 1.224 m3/s, 2594 CFM, tile flow = 0 m3/s,0 CFM, 0% of rack flow).

    Velocity

    (m/s)

    (a) PIV (b) Porous jump (c) Body force (d) Modified body

    force

    (e) Geometrical

    resolution

    Figure 6. Comparison of PIV and CFD results for case 2 (rack flow = 1.224 m3/s, 2594 CFM, tile flow = 0.234

    m3/s, 496 CFM, 60% of rack flow).

  • 8/13/2019 2012 - Rack Level Modeling of Air Flow Through Perforated Tile in a Data Center

    11/11

    11 Copyright 2012 by ASME

    Velocity

    (m/s)

    (a) PIV (b) Porous jump

    model

    (c) Body force

    model

    (d) Modified body

    force

    (d) Geometrical

    resolution

    Figure 7. Comparison of PIV and CFD results for case 3 (rack flow = 1.224 m3/s, 2594 CFM, tile flow = 0.754m

    3/s, 1598 CFM, 60% of rack flow).

    Velocity

    (m/s)

    (a) PIV (b) Porous jump (c) Body force (d) Modified body

    force

    (e) Geometrical

    resolution

    Figure 8. Comparison of PIV and CFD results for case 4 (rack flow = 1.224 m3/s, 2594 CFM, tile flow = 1.224

    m3/s, 2594 CFM, 100% of rack flow).