Data Center Cooling

One of the most critical parts in designing a data centre is to calculate the optimum amount of cooling power required for keeping the IT equipment running and also to avoid over cooling spending energy without necessity. In order to calculate the cooling load the following methodology was used:

P_cooling is the cooling load per rack in MW/rack.

P_server is the total power of the facility running at 100% of the load, but also in this work an expansion plan of 50% was also account, given the following value:

With this expansion plan it would be possible to change the power density of the racks without requiring major work on adapting the cooling system, at 5MW the cooling system wouldn’t run at full load, increasing as the power density raises.

N°_of_Racks is the total number of racks in the installation previously calculated on the datacentre dimensions section.

P_Heat gains is the total heat gain from lights, people and fabric (walls and roof).

In order to calculate the total heat gain a Simulink was built, to simulate a 24 hours of a worst case scenario in which the outside temperature is the highest recorded in 2014. The Simulink model is described in Figure 1.

Figure 1 - server room Simulink model

(1)- Heat gains from people of 100W per person [1] that varies with the occupancy of the server room.

(2)- Heat gains from lights of 7W per LED that is 70% of total LED power [2], the total heat from light also varies with the number of occupants in the data centre, the function of occupancy in the server room can be seen in the lighting section of this website.

(3)- Heat gains from the servers, in this case a total of 7.5MW as explained previously.

(4)-Simulates a heat flow through a constant temperature of 300K

(5)- is the heat flow through the fabric (wall and roof), both present a of K=0.04 W/m.K and a thickness of 400mm, these thickness and insulation levels are more than enough to avoid condensation as follows:

Table 1 - minimum thickness required

(6)- is the outside temperature of the hottest day in 2014 (day 170 on 19/06/2014)

(7)- is the final result that divides the total power by the number of racks.

Note in Figure 2 the peak value is during the mid day when the outside temperature is the highest, giving a value of 6,1 kW/rack of cooling load in worst case scenario.

Figure 2 - Cooling power require per rack

Technology selection

Before selecting the cooling technology it is necessary to know what are the standards and regulations for the inside air the datacentre. In this work was used the ASHRAE recommendation as can be seen below:

The aim of controlling humidity and temperature is to maintain the air inside the datacentre between the recommended envelope for classes A1-A4, by keeping it inside the standards it is possible to run the IT equipment without any risk of overheating (when the server reaches a temperature which activates a self-shut down to avoid data losses) or static discharges that happen when the inside air is too dry also causing a data loss.

Now that we have defined inside air properties, a psychometric analysis can be done to determine which of cooling or heating technologies has to be used to keep the outside air in ASHRAE standards, in order to do this analysis the following model was used as base:

Two Matlab algorithms were used to analyse the ambient air properties of Falkirk, based on an hourly updated table from air properties of OBAN a city with a similar climate to Falkirk. The first algorithm has as inputs the temperature (°C) and humidity ratio (g water/kg of dry air) and as output a psychometric chart with the air properties for the entire year, the read area is the recommended temperature and humidity, as shown below:

The second algorithm has for input the air temperature (°F), relative humidity (%), enthalpy (BTU/lb) and humidity ratio (lb Water/lb Dry air) and for output the total number of hours required technology to keep the air in the ASHRAE standards, as shown in the Table 2 and Table 3:

Both algorithm can be found on the software section of this web site.

By analysing the Table 1 a system with direct evaporation + economizer with a backup system with DX (direct expansion or mechanical cooling) which will allow the use of 99% of free cooling as selected, the main challenge is to quantify how many coolers will be required.

Table 2 - free cooling available during the year

Figure 5 - Psychometric Bin Analysis for Falkirk

Figure 4 - Psychometric Bin Analysis for Alternative cooling [3]

Figure 3 - ASHRAE recommendations

Table 3 - free cooling available during the year

Manufacturer Selection

Now that the air flow (see the ventilaton section for more details) and cooling power per rack was calculated, it is possible to quantify the number of coolers and humidifiers, manufacturers and quantity of coolers can be seen below.

References

[1] engineering toolbox, "People and Heat Gain," [Online]. Available: http://www.engineeringtoolbox.com/persons-heat-gain-d_242.html. [Accessed 11 05 2015].

[2] mechatronix-asia, "LED heat sink calculation," 2015. [Online]. Available: http://www.mechatronix-asia.com/LED_heat_sink_calculation_simulation_thermal_design.html. [Accessed 05 2015].

[3] O. V. P. C. R. P. J. D. a. C. K. Ian Metzger, "Psychrometric Bin Analysis for Alternative Cooling Strategies in Data Centers".