1、32 2010 ASHRAEINTRODUCTIONA key focus in the IT market continues to be on coolingstrategies and heat removal from medium to high-density heatloads. Whether at the component, enclosure or room level, endusers are struggling to keep up with constantly escalating heatloads. And a condition being addres
2、sed almost as rapidly asthe heat buildup, with a wide variety of solutions, configura-tions and components, either available now or in the nearfuture, that will help, at the least, reduce the impact of havingtoo much heat to get rid of.Every facility operatorsmall to large, end user to host-ing site
3、, private or public sector, is aware of this situation andtrying to cope as best as possible. Many strategies have alreadybeen successfully implemented: Use of rack mount blankpanels, vertical air baffles, improved floor tile management,hole grommets, CRAC/CRAH unit placement, etc. are all incurrent
4、 use to maximize facility climate control capacity.Improved cable management and control and use of underfloor and overhead ceiling spaces can also stretch theseresources. Even the revised ASHRAE specifications for oper-ating conditions can increase heat removal capacity. But onlyup to a point.And a
5、t many locations, that point has been reached.With these greater loads and higher density installationsbecoming more commonplace, two newer methods are beingdeployedaisle containment solutions and close-coupledheat transfer systems. Both product sets are designed toincrease heat removal capacity in
6、the IT space. However, theIT communityfrom the design engineer, through compo-nent vendor and ultimately the end user, are tasked to not onlydesign, install and operate a solution but to also provide thedataoperational performance results, to justify the expenseand verify the claims. This has led to
7、 a new awareness of bestpractices for data centers, with the clear goal to improve facil-ity operations and performance. This situation has alsoexpended a great deal of planning resources to insure IT facil-ities will be able to handle future growth and new productdeployments.This paper will review
8、some of the design considerationsfor a Cold Aisle Containment (CAC) installation and opera-tional capacities that can be achieved with the system.System ReviewSeparate the air. This has emerged as a key strategy inimproving cooling and heat removal capacities in IT facilities.In addition to enclosur
9、e (blank panels, vertical baffles) andfloor solutions (tile management, hole grommets), a next stepto be considered would be to enclose the actual hot or coldaisle spaces within a facility. This type of installation wouldsignificantly improve airflow management to active compo-nents, providing maxim
10、um separation between the hot andcold airflow paths.Any aisle containment system will consist of just a fewcomponents that should be designed to provide easy retrofitand deployment into existing spaces and onto existing enclo-sures. The two key elements will be row end doors to close offaisles and c
11、eiling panels to cap aisle spaces. Whether choosingto seal off the hot aisle or cold aisle, these systems should beable to be placed in either raised or solid floor installations andwork in conjunction with existing facility climate controlcomponents. And if being considered for new construction orf
12、or use in a facility that has reached maximum heat removalcapacity, an aisle containment solution should work with inHigh Density Cooling SolutionsTaking IT to the Next Level: The Cold Aisle Containment AlternativeHerb VillaAssociate Member ASHRAEHerb Villa is a customer solutions engineer with Swit
13、ch and Data, Tampa, FL.OR-10-005 2010, American Society of Heating, Refrigerating and Air-Conditioning Engineers, Inc. (www.ashrae.org). Published in ASHRAE Transactions 2010, Vol. 116, Part 1. For personal use only. Additional reproduction, distribution, or transmission in either print or digital f
14、orm is not permitted without ASHRAEs prior written permission. ASHRAE Transactions 33row cooling systems to provide additional heat removal capac-ities (see Figures 1 and 2).Test ReviewTesting of the Cold Aisle Containment (CAC) system wasconducted in the Dallas-Metro region of Texas in an opera-tio
15、nal data center with a 24 in./610 mm raised plenum floor. Aseparate test area, approximately 1100 ft2/102.2 m2, was iso-lated from the main data center by using insulated walls placedfrom the subfloor, through the raised floor and up to the dropceilingpreventing any air from mixing with main data ce
16、nterspaces. The test setup consisted of 10 server enclosures placedin two facing rows of 5 enclosures each with a four-foot coldaisle between them. The individual enclosures were 24 in./600 mm wide and 48 in./1200 mm deep, providing 42U ofmounting space, and each one was isolated from the adjacenten
17、closures with an internal partition wall. Row ends weresecured with sidewalls. The enclosures were configured witha standard 64% perforated single door on the front, split per-forated rear doors, and a perforated roof. Vertical air baffleswere installed on the front 19 in. component mounting rails t
18、oensure proper airflow and prevent any internal cabinet airrecirculation and mixing.The cold aisle was fully contained on the aisle ends with180-hinged double doors. The top of the aisle was coveredand sealed with clear polycarbonate panels. Ten perforatedfloor tiles were installed in the contained
19、cold aisle to providecold air supply from the raised floor plenum. Numerous testswere performed utilizing perforated tiles with 25, 65, and 80%open surface area. Nine (9) 2.4 kW load banks were installedin each test enclosure, each occupying 4U of space, whichallowed for a test range up to 20 kW per
20、 enclosure, or 200 kWfor the entire contained area. The load banks were designed tomimic server airflow patterns with a nominal T (temperaturechange) of 45F/7.2C (as expected with newer, high densityservers) and with fans that could deliver 300 cfm per loadbank.Three (3) 20-ton nominal, direct expan
21、sion CRAC unitsprovided space cooling. Power was provided from a 225 KVAPower Distribution Unit (PDU). All equipment and ambientroom spaces were instrumented, providing a complete array oftemperature and environmental sensors to ensure accuratedata collection.Test ScenariosDuring testing, steady sta
22、te comparisons for the systemwere measured to determine operational capacity. Tests werecompleted with 25 and 65% perforated floor tiles. 80% perfo-rated floor tiles were also tested. cfm data for each tile con-figuration was recorded to determine the total airflow in theCAC area. Average inlet air
23、temperatures, outlet air tempera-tures, room ambient temperatures and hot aisle temperatureswere recorded to capture steady state operation for all threescenarios.During the test, a maximum allowable T of 45F/7.2Cwas established. Since newer, high-density servers have muchhigher outlet air temperatu
24、res than older servers, testing withthe higher is desirable. This can be illustrated simply bycomparing a variety of new 1U and blade server chassis, whichcan have T s as high as 52F/11.1C (Dell 2008, CapacityPlanner).65% Perforated tiles were chosen as an operational stan-dard. The majority of test
25、ing conducted utilized 65% tiles.Each enclosure was set to a 20 kW load with all 3 CRAC unitsoperated at full capacity. Each CRAC unit brought all of itscompressors online to handle the maximum load, with steadystate operation achieved after approximately one hour. Peaktemperature increases were not
26、 noted until the full thermalcapacities of the CRAC units were reached, at which timethe load was reduced. The average T was approximatelyFigure 1 Aisle containment with raised floor and CRAC.Figure 2 Aisle containment with solid floor and in rowcooling. 2010, American Society of Heating, Refrigerat
27、ing and Air-Conditioning Engineers, Inc. (www.ashrae.org). Published in ASHRAE Transactions 2010, Vol. 116, Part 1. For personal use only. Additional reproduction, distribution, or transmission in either print or digital form is not permitted without ASHRAEs prior written permission. 34 ASHRAE Trans
28、actions30F (1.1C), not the worst-case scenario of 45.0F (7.2C)as described earlier. Additional Operational ScenariosAs any IT professional can attest, a data center is a verydynamic and ever changing environment. And while the initialtest conditions were valuable, it was decided to view systemperfor
29、mance under a variety of operational scenarios that occurin the IT space. And even under adverse or abnormal conditionsthat can occur due to component or resource failure.Tests started with a “typical” installation, simulating aspace not built with any means to separate and/or contain thecold or hot
30、 air streams in data center spaces. Recirculation andmixing occur, reducing the overall effectiveness of themechanical system to reject the heat from IT loads. Compar-ing data with containment to data without containment wasone goal of the test and considered necessary to determine thetrue load capa
31、city of the mechanical cooling system. Thiscomparative data will help illustrate the gains achieved byusing a cold aisle containment system.To demonstrate this, the doors and roof panels of the coldaisle containment system were removed and the aisle wasfitted with standard 65% perforated tiles. The
32、system wasstarted with a single CRAC unit operating, with additionalCRACs brought online to prevent server temperaturerunaway as additional load was added.As it turned out, for all testing above 6.9 kW per enclo-sure, the T observed was found to be unacceptable. Evenassuming high T servers, the maxi
33、mum load achieved duringtesting without containment was 11.5 kW per enclosure. Themaximum load was determined when the load banks went intothermal overload, much like a typical server would. It is clearfrom the data that cold aisle containment resulted in signifi-cant performance improvements over t
34、he non-contained aisletesting. This is assumed to be a result of the complete elimi-nation of air mixing and the slight pressurization of the coldaisle when the containment system was utilized.And then what happens when something goes wrong?Typical data centers operate with some level of facility, s
35、ystemand component redundancy. The ability to overcome unfore-seen failure events is critical for any data center. During test-ing, a number of failure scenarios were simulated to determinethe impact of the cold aisle containment on different faultcondition scenarios.Generator Startup / UPS Fail Ove
36、rTo replicate this failure, a load of 20 kW in each of 10enclosures was established with all CRAC units running. Atexactly the same instant, all of the CRAC units were de-ener-gized at their disconnects, resulting in the total loss of air flowto the cold aisle containment system. After approximately
37、 30seconds (to simulate emergency generator start up) the CRACunits were restarted and began to supply air to the equipment.Power requirements of the load were monitored at all times toensure that no single load bank was lost on a thermal overload.The generator restart resulted in a temporary increa
38、se in thehot aisle temperature, as was expected due to the loss ofairflow. The shutdown also impacted the cold air temperature;undoubtedly due to the extra thermal energy that the CRACunits were required to dissipate after restart.Opening of Aisle End DoorsPersonnel activity is a day-to-day reality
39、in a data center.Unfortunately, these movements can disturb the air paths thatare required to maintain a sealed cold aisle containment system.These impacts are usually momentary, but what if a row enddoor of the cold aisle containment space was opened for over 30minutes when running at 20 kW per enc
40、losure? Room ambienttemperature, as well as hot aisle temperature increased due tothe short cycling of air from the cold aisle back to the CRACunits. This did reduce cold aisle temperature due to the lowertemperature of air being cycled directly to the CRAC units.Even in this state, however, the air
41、 temperature to the serverswas maintained within the normal operating range, andalthough the T rose due to lower air volumes being providedto the servers, the T was still within established guidelines.Loss of Single CRAC UnitThe system that was tested could only be considered an“N” redundant solutio
42、n since all CRAC units were required tobe operational to support the 20 kW/enclosure IT test load.Though many data centers operate in at least an N+1 redun-dant configuration, what if this werent the case? During test-ing, the system was operated at a steady state of 20 kW/enclosure, and then a sing
43、le CRAC unit was shut down. TIncreased quite rapidly, and inlet temperature to the load bankswas affected over time. The two remaining CRAC units wererated for a theoretical 70 kW capacity each, but with the lossof one of the CRAC units, the remaining two units were ableto support 100 kW each. By th
44、e end of the test, it was deter-mined that the load was indeed stable at these conditions,allowing sufficient time for repair of the “failed” CRAC. Benefits Review: CapEx and OpExThe data gathered is valuable in understanding the oper-ation of a cold aisle containment system. However, it is critical
45、to understand potential operational and installation CapEx(Capital Expenditure) and OpEx (Operational Expenditure)cost savings that can be realized from a relatively small coldaisle containment system.The operational efficiencies data seen in Table 1 showthat, on average, when considering a load of
46、10 kW/enclosure,a single 20-ton CRAC unit has the effective capacity of 50 kWwhen used without containment and a capacity of 70 kW perunit with containment. This improvement results in a reduc-tion of the total CRAC units required Day 1, thereby reducingCapEx costs and reducing overall costs through
47、out the year byvirtue of lower OpEx. 2010, American Society of Heating, Refrigerating and Air-Conditioning Engineers, Inc. (www.ashrae.org). Published in ASHRAE Transactions 2010, Vol. 116, Part 1. For personal use only. Additional reproduction, distribution, or transmission in either print or digit
48、al form is not permitted without ASHRAEs prior written permission. ASHRAE Transactions 35CONCLUSIONA key strategy to increase cooling efficiency that iscurrently being deployed in operational data centers anddiscussed by the IT market at large, is to maximize the sepa-ration of cold and hot air flow
49、 paths. Aisle containment hasemerged as a relatively low cost and easy to implement solu-tion. The testing of the Cold Aisle Containment system vali-dated the assumption that utilizing a cold aisle containmentsystem enables the data center environment to support higherdensity loads while providing a more efficient use of the envi-ronmental equipment. Increased efficiencies of up to 40%were achieved utilizing cold aisle containment in comparisonto environments without containment. The CRAC units wereable to support Ts (supply/return) ranging from 75 to
