1、CH-06-9-2 , An Analysis of the Effects of Ceiling Height on Air Distribution in Data Centers Vali Sorell, PE Member ASHRAE Yousef Abougabal, PE Associate Member ASHRAE Viralkumar Gandhi Associate Member ASHRAE ABSTRACT In this study, computationalfluid dynamics (CFD) models are used to evaluate how
2、ceiling height afects the overall performance of the air distribution system within a data center environment. First, an underfloor air delivery system is exam- ined while keeping all operational parameters constant and varying only the clear space height. Then, the performance of the same system is
3、 evaluated with a dropped ceiling while vary- ing the clear space height. Finally, an overhead air distribution system is considered, keeping all operational parameters constant while varying only the clear space height. Theperfor- mance of each system is evaluated using dimensionlessparam- eters de
4、veloped previously by others. These indices, supply heat index (SHr) and return heat index (RHI), when used to examine the temperatures at the inlets to the equipment inside the cabinets and at the air handler or computer room air- conditioning unit (CRAC) inlets, provide a measure of how much recir
5、culation and bypass occur within the data space. INTRODUCTION Users and designers of data centers and telecommunica- tions equipment rooms have often debated the relative merits of providing a ceiling over the equipment space. Sometimes a ceiling is provided for acoustical or aesthetic purposes alon
6、e; at other times the ceiling is provided in order to use the space above the ceiling as a path for the return air. Recent practice for many new data spaces has been to use the ceiling as a return air plenum; this has been accomplished by extending a return duct from the inlets of the computer room
7、air-conditioning (CRAC) units to the ceiling. This appears to follow the gener- ally accepted notion that using the ceiling plenum as an active participant in the return air system improves the overall perfor- Kishor Khankari, PhD Member ASHRAE Aas h is h Watve Associate Member ASHRAE mance of the d
8、ata space by placing a physical barrier between the hot and cold airstreams. Although this seems intuitively correct, there are no published data to support this approach. In fact, some have even questioned whether a ceiling is neces- sary. The question raised is whether it is possible to provide en
9、ough free area in the ceiling directly above the hot aisles to allow the capture of all of the hot aisle air into the return plenum without spillover (or recirculation) into the cold aisle. If there is recirculation, one needs to ask how many more ceil- ing tiles need to be removed in order to avoid
10、 recirculation. And if more tiles are removed, is there still value in having a ceiling at all? Another important consideration in the development of a data space is the overall height of the space. Obviously, many designs are limited to the clear heights available within the space designated to be
11、converted to a data space. Usually, space height is used as a selection criterion only to the extent necessary to ensure that equipment cabinets, IT cable manage- ment systems, lighting, ductwork (if used), and a raised floor system (if used) can fit within the space. Likewise, when new construction
12、 is pursued, the design for the height of the build- ing usually considers only that all the required equipment and systems fit into the space. In both cases, consideration is usually not likely to be given to optimizing the air distribution system. Although there may be a substantial cost associate
13、d with raising the roof several feet, such cost should be assessed considering that a taller space may improve the airflow dynamics of the space, thus allowing servers or equipment to be placed in cabinets where the environmental conditions would otherwise not have been favorable. This study examine
14、s these space and ceiling height parameters to address two important site selection andor Vali Sorell and Yousef Abougabal are with Syska Hennessy Group, Inc., San Francisco, Calif. Kishor Khankari is lead consulting engineer with Fluent, Inc., Ann Arbor, Mich. Viralkumar Gandhi and Aashish Watve ar
15、e with Fluent Inc., Lebanon, NH. 02006 ASHRAE. 623 design issues: (1) the optimal conditions for the use ofa ceiling as a return air plenum and (2) the benefits of selecting as tall a space as possible. These parameters are examined by applying indices developed by Sharma et al. (2002). The indices
16、provide a means of evaluating how much hot air recirculates from the hot aisle into the cold aisle and how much cold air bypasses the cold aisle to the return at the CRAC unit. Applying these indi- ces using computational fluid dynamics (CFD) modeling allows one to evaluate the performance of variou
17、s designs before they are built. The CFD models enable the designer to test alternative designs by extracting any number of relevant temperatures from the domain space of each tested design. SPACE HEIGHT%, FT(MJ ALTERNATIVE DATA CENTER DESIGNS UNDERFLOOR WITHOUTCEIUNG UNDERFLOOR WITH CEILING OVERHEA
18、D WITHOUT CEIUNG TRIALWI TRIAI 87 TRIAL93 TRIAL #I TRIAL U7 TRIAL83 TRIALWI TRIALU2 TRIALff3 120 140 160 I20 140 160 120 140 160 (366) (427) (4861 (366) (427) (468) (366) (427) (488) The Starting Point-“A Shell Space“ The mechanical design criteria for the modeled spaces were set similar to designs
19、currently found in many data spaces. The overall load density was set to 100 W/ft2 (1076 W/m2) in a space of approximately 9,540 ft2 (886 m2). The space is 159 fi (48.5 m) long and 60 fi (1 8.2 m) wide. The ceiling height (where used) and the roof height were kept as the variable. Where a raised flo
20、or was used, the raised floor height was set to 30 in. (0.76 m). The model used for this study is based on one used for a previous study that examined the overall performance of two air distribution methodologies as a function of total airflow (Sorell et al. 2005); that model is used here with some
21、slight modifications. In the interest of conserving space, the basic description of that model is simplified herein. CASE A UNDERFLOOR AIR DISTRIBUTION WITHOUT CEILING Three different configurations were developed within the CFD model. Each is summarized in Figure 1 and described in further detail a
22、s follows. Case A: Underfloor Air Distribution without Ceiling With the basic load and spatial parameters determined, the shell space was populated with servers, PDUs, and CRAC units such that the load over the entire space equated 100 W/ft2 (1076 W/m2). The equipment cabinets were modeled as though
23、 the flow across the perforated doors from front to back experiences a uniform airflow equivalent to a temperature rise of25“F (14C). This corresponds to an equip- ment airflow of approximately 120,000 cfm (56 m3/s) across the entire data space. Fifteen CRAC units were distributed along the perimete
24、r space, 13 of which were modeled as ON. (This represents the realistic condition in which two redundant units are kept in standby mode and consequently allows for a small amount of leakage from the underfloor plenum to the data space.) The total airflow through the 13 active CRAC units represents 1
25、10% of the total equipment airflow (roughly 132,000 cfm 62.3 m3s). This assumption of 10% more supply air than equipment air represents a workable airflow for both overhead and underfloor air delivery systems given these current loads, temperatures, cabinet configurations, and system arrangements (S
26、orell et al. 2005). The airflow rate was set such that all active units provided the same flow, and the supply air temperature was set uniformly to 60F (1 5.5“C). The rows of equipment cabinets were arranged in a hot- aisle/cold-aisle arrangement with perforated floor tiles located in alternating ai
27、sles. The raised floor was set to 30 in. (0.76 m) high. A return duct extension was added to take return air from a level 30 in. (0.76 m) below the bottom of the roof deck. This follows the common practice of taking return air from where it is anticipated to be representative of the hot aisle discha
28、rge. CASE B UNDERFLOOR AIR DISTRIBUTION WITH CEIUNG CASE C OVERHEAD AIR DISTRIBUTION WITHOUT CEILING SUPPLY GRILLE (OVERCOLDAISLEJ SUPPLY DUCT EOUIPMENT Oi 2 WNEIS 9 SPACE HEIGHT CASE A I CASE B I CASE C Figure 1 Cases A, B, and C dejned. 624 ASHRAE Transactions: Symposia Figure 2 Cases A and B: und
29、erfloor air distribution. Figure 3 Case C: overhead air distribution. Three trials were examined for this case in which the space height (X) was varied from 12 to 16 feet (3.66 to 4.88 meters) in 2-foot (0.61 -meter) increments. Figure 2 represents this case as seen within the CFD model. Case B: Und
30、erfloor Air Distribution With Ceiling room. However, just as in Cases A and B, Case C assumed a total equipment load of 100 W/ft2 (1076 W/m2). Therefore, although the airflow through individual cabinets may vary slightly between each model, the total load, the total airflow, and the equipment AT are
31、 the same through all three cases. Figure 3 represents this case as seen in the CFD model. All parameters of Case A were duplicated in Case B with the only difference being that a ceiling with return grilles was added at a level 30 in. (0.76 m) below the roof deck. With this arrangement, the return
32、duct extensions extract air directly from the ceiling plenum. The return grilles are arranged directly above the hot aisles with sufficient coverage such that the size of each return air opening corresponds to the size of each hot aisle. Case C: Overhead Air Distribution Without Ceiling Because an o
33、verhead system needs to be ducted, it would be expected that the ability to deliver higher operating pres- sures would be greater. Consequently, the basic assumptions for the air delivery systems were modified to model an arrangement that would be more commonly found for such an air distribution sys
34、tem. Nine built-up, upflow air-handling units (AHUs) were modeled for this case, and since these systems can be typically provided with variable frequency drives (VFDs), all nine units were modeled as ON with each delivering approximately 14,700 cfm (6.7m3/s) for a total of 132,000 cfm (62.3 m3/s).
35、As in Cases A and B, the supply air temperature was set uniformly to 60F (15.5“C). Also, being upflow units, the return air inlets are at the low side ofthe unit. Therefore, a 9-foot-high (2.74-meter-high) partition wall was added around the units to force the intake of the warmer hot aisle air to e
36、nter the AHUs from above (or, stated differently, the intent was to prevent the cold aisle air from migrating out of the cold aisle directly into the AHU inlets). The layout of the rows and equipment cabinets is slightly different between this model and the underfloor models due to the different typ
37、e of fan systems required and the need for a fan As in the previous cases, three trials were examined with the space heights taken at 12, 14, and 16 ft (3.66, 4.27, and 4.88 m). In all trials, the supply grilles were located directly over the cold aisles at an elevation of 8.5 ft (2.60 m). The inten
38、t of creating a CFD model of a ducted air distri- bution design was not to model the airflow within the duct- work. Since it can be safely assumed that a ducted system can, if designed properly, be balanced to deliver any reasonable air quantity from any outlet, each outlet was modeled as a fan deli
39、vering the desired quantity of air at the appropriate loca- tions above the cold aisles. This does not mean that the ducts themselves were not included in the analysis. In fact, every duct was sized and every outlet was selected according to best engineering practices. Although the airflow inside th
40、e ducts was not modeled, the effect of the relatively large ductwork within the space and the outlet jets over the cold aisles were expected to have significant bearing on the final outcome of the analysis. METHOD FOR QUANTIFYING THE PERFORMANCE OF EACH MODEL Once the three models are created and te
41、sted within the framework of CFD, the problem still remains of how to eval- uate that one trial is more or less effective than the others. Designers often develop CFD modeling schemes for data spaces that look at underfloor pressure and/or air velocities, presumably to equate a uniform underfloor pr
42、essure distribu- tion with the ability to deliver a uniform air distribution through perforated tiles. Not only is this method not appropriate for a ducted system, but delivery of air through perforated tiles (or through difisers for the ducted option) does not tell the full ASHRAE Transactions: Sym
43、posia 625 story of whether the cold air actually arrives where it counts most-that is, at the inlet of the servers or cabinets. CFD models have also been created in which temperature profiles are generated to demonstrate that the cold aisles are uniformly cold and the hot aisles are uniformly hot. T
44、his approach is more meaningful since it gives more of an indica- tion of whether the environment from which servers and cabi- nets extract air is appropriate for cooling the critical equipment. Unfortunately, this approach is at best qualitative and tells nothing of the overall effectiveness of the
45、 air distri- bution system established within the space. Sharma et al. (2002) developed a method of quantifying the performance of a data center by looking at two closely related dimensionless indices-the supply heat index (SHI) and the return heat index (RHI). Use of Supply Heat Index and Return He
46、at Index The SHI is defined as an enthalpy rise due to inJiltration in the cold aisle from the supply air outlet to the cabinet inlet divided by the total enthalpy rise from the supply air outlet to the cabinet outlet. or SHI = SQ/(Q + SQ), where SQ represents the enthalpy rise of the supply air due
47、 to infiltration of hot air into the cold aisle and Q represents the heat rejectcd from the cabinet or rack equipment. This can be expressed as where mw Cp = the specific heat of air, T, = temperature of the air entering the cabinet, Tsupplr = supply air temperature, and Tout = temperature of the ai
48、r leaving the cabinet. = the mass airflow rate across all the cabinets where the temperatures are sampled, For this study the assumption was made that the mass flow rate across the servers is uniform across all the sampling points at the inlets and outlets to the servers. Therefore, the mijCp terms
49、cancel and the SHI can be calculated across all the sampling points as Similarly, the RHI is defined as the total heat extraction by the air-handling units divided by the total enthalpy rise from the supply air outlet to the cabinet outlet, or RHI = Q/(Q + SQ). This can be expressed as where hk = the mass airflow rate across the CRAC units (or AHs) where the temperatures are sampled and Tretum = temperature of the air entering the CRAC unit (or AWs). For this study the assumption was made that the mass flow rate across the CRAC units (or AHUs) is 110% of the mass flow rate across the
