Boosting the Energy efficiency of a Server Room: experimental + virtual approach. Nowadays, the increase in computing power allows the successful applications of the CFD simulation methods in civil environments. This approach enables the evaluation of the energy performance in an civil environment and, subsequently, to assess the impact of changes aimed to increasing efficiency. Prompt Engineering and LDV were committed by a well -known European banking group with headquarters in Italy, in the Server ro om performance objectification. In addition it was requested to predict, numerically, the Datacenter performance after a partial refurbishment. THE BACKGROUND The design of a energy efficient data center or server room requires attention, capacity and investment, but if it is done properly provides benefits during the whole room lifetime (reducing the running cost and the management). The hardware s into a data center are many; each of those must not be chosen independently, but as a tile in the whole jigsaw. It is crucial to properly assess the internal energy flow, electrical and thermal power. The three main factor to be kept in account during the design of a datacenter are: Hardware reliability / redundancy HW scalability and modularity HW efficiency These features must be taken into account in the selection of all the components related to the data center design, whether IT, power suppliers or conditioning plant. Prompt Engineering and LDV have proposed to the committer a numerical and experimental approach, aimed initially to have an accurate snapshot of the server room energy performance into its original configuration (ANALYSIS - layout, consumption). Then, using its expertise in the field of CFD, PE & LDV have developed a 3d calibrated numerical model of the server room, focused to replicate the measures collected in the previous phase (NUMERICAL MODEL). The server room numerical model was subsequently made available to the data center design expert, allowing them to have an accurate predictive tool; each solution they ve proposed has been replicated and numerically evaluated in the typical parameters for energy efficiency. ANALYSIS (Energy consumption, layout, average running temperatures) One of the fundamental operations to achieve an high energy efficiency data center is the power consumption measurement (or estimation) for each equipment included into the server room. This is valid both in the case of the data center in the design phase, either in the case of renovation or expansion (as in the present case). The power consumption of a data center can be summarized in three main contributions: the power consumption of the equipment (50%), consumption for cooling (30%) and consumption for the IT stuffs "active work" (20%). The analysis
took into account the efficiency of energy consumption for related to the cooling. The scenario where PE and LDV were requested to work, there aren t energy consumption data available. Moreover, there aren t available temperature data related to the server room. They were fully available geometrical data and the datasheet of each individual unit included in the server room. 3: AC data / scheme 1: scheme and nomenclature The equipment in the server room (racks) are arranged to create hot and cold aisles. The hot air coming out from the racks enters the air conditioners (CRAC), positioned in the hall, where it is cooled. The conditioned air is conveyed below the floor (under-floor plenum) and returns back to the room though the cooling grids into the racks. The corridor where the grills are, is named cold aisles. The data correlation from the numerical data and the experimental results it is crucial to built a strong numerical model. To do so, specific sensors were scattered in the room in order to collect the needed data. 4: Sensors position Those data have been collected by a 24h continuous logging. Moreover, in order to have more data and field information on the temperatures, IR maps have been taken. Those will give more details on the temperature distribution on the hot component in the room. The temperature maps are a focal point for the correlation. 2: server room layout (CAD) The operating data of machines installed into the server room (air conditioners and the IT stuffs) were obtained from a datasheets.
The venting grid on the floor have been modeled accurately, because of they represent the section in which the cold air flow is coming into the server room. The steady state simulation allows the comparison between the numerical temperature map of each rack and the experimental campaign data taken before. 5: Thermographic (IR) image -example NUMERICAL MODEL and SIMULATION Thanks to the geometrical data available and to the information collected, a numerical FEM model has been created. The model is based on a structured mesh, brick made. The model kept in account the constructive material for each model part (floor, racks, walls, ) 7: Simulation vs experimental results The field data taken from the sensors are compared with the same information coming from the numerical model. 6: Simulation loop Every single rack in the server room has been represented in the model by using a normalized electrical power consumption (0-25% Pmax, 25-50% Pmax, 50-75% Pmax, 75-100% Pmax).. That info is collected by checking each rack datasheet. In the same time, every Air conditioner has been modelled by taking the datasheet info as well as the geometrical representation of the inlet and outlet sections.
30 25 20 15 10 5 0 I27 S2 S27 S12 S21 I15 S15 S23 I28 S28 S19 S3 I31 S31 I1 S1 S17 S26 I26 S30 S18 S9 I3 I28b S32 S8 I19 S4 S31b S4 8: field data: simulation temperature distribution @ rack level. Below the comparison between virtual and real temperature sensors The numerical and the test results well fits; so the numerical model created is validated and predictive in the strategic area selected as important. The model as it is now is well representative of the physic of the server room. The model has been delivered to the expert to enable them to make virtual experimentations aimed to boost the datacenter efficiency. WHAT IF ANALYSIS IT specialists decide to analyze two different configurations. Each of those have been replicated in the numerical model and used to run a simulation. The goodness of the modification or, even better the energy efficiency performance objectivation has been done, comparing specific phenomena before and after the modification. I.g. the reduction of the mixing in between cold and hot air in the suction section of the CRACs or the cold air short circuits close to the racks. In the literature, specific indexes such as the SHI (Supplied heat index) are used to make have a performance objectivation The modifications included in the what / if analysis are: a) Grid vents extention and orientation change mean numerico b) Power reduction achieved by a selective CRAC shutdown. This strategy should be applied in the server room region in which the temperature is not critical. Anyhow the redundancy mainstream useful in the server room design has been accomplished. Both configurations have been simulated starting from the predictive numerical model as per the chapter before. The on the floor venting grid modification implies a geometrical change on the numerical model itself. The analysis results brings to the conclusion that the change is modifying the fluid dynamics in the plenum below the floor. Being the server room a closed system where the flow rate blow by the CRAC is constant, the air flow distribution across the venting grid is not intuitive and easily predictable. Simulations shows the fluid dynamics below the floor and the experts were able to optimize the air flow by operating on the grid areas and orientation. The CRAC power reduction has been simulated by changing some parameters in the the CRAC numerical model. Even if each CRAC in the server room is designed accordingly with the redundancy approach, the analysis performed brings to the conclusions that some parts of the room itself are overcooled the coolant power density is too high. Those peaks, where not useful to keep cool the IT stuff, are useless and against the energy efficiency. In order to evaluate the effect of the two modification coupled in the server room, a specific model has been set up. The analysis of this new model allows to find the right balance in
between the energy efficiency contribute of each modifications. CONCLUSIONS The methods of numerical simulation are universal and can be used with success in nontraditional contexts. The buyer has positively affected by the turning point in that: the restructuring of the server room has been performed by having a clear idea of the impact of any change on the energy efficiency of the server room taking in consideration the costs associated with each change, it is possible to evaluate a complete business case concerning the energy efficiency and the return on investment Once in operation, the data center has presented much lower operating costs