Well said. The breakdown occurs when you look at the airflow required for higher watt density equipment such as blade servers. You could theoretically fill a std. 42U cabinet with 6 of these 7U chassis. Seeing that each chassis (using IBM here as an example) can pull 4kW of power, we could draw 24kW of power per cabinet. In most data center layouts, cabinets generally have one perforated tile in front of them which can supply 300 cubic feet per minute (CFM) based on std. design under-floor static pressure. To maintain an acceptable temperature rise through the server (so the exhaust isn't melting things), we need to supply approx. 120 CFM per kW of load. For 24kW that would be 2880 CFM. Seeing that the cabinet only has one 300 CFM tile in front of it, we have a problem Houston. There are variety of successful air-side strategies to deal with this including supplemental air delivery and exhaust air segregation. Ultimately, to go much higher than 24kW, we may need to use liquid cooling, which has the ability to remove far more heat, but with the liabilities mentioned by others. Cooling the room more doesn't help, as it's really an air delivery issue, regardless of supply temperature. In fact, overcooling is tremendously expensive due to the dehumidification effect of really cold cooling coils. The humidity then has to returned to the air to maintain the 50% RH goal. There are huge energy losses associated with this. Additionally, when the under-floor supply temperature goes below the dew point, some surfaces will as well, and they will have condensation occurring on them. Needless to say, condensation is not good for PC boards. In the short term, separation of the supply air and the exhaust air to avoid mixing is a good strategy. By avoiding mixing, assuming the available air volume is sufficient, high density power consumption can be cooled. A raised computer floor isn't even necessary for this if the air delivery method is designed right.
Slashdot Mirror
Well said. The breakdown occurs when you look at the airflow required for higher watt density equipment such as blade servers. You could theoretically fill a std. 42U cabinet with 6 of these 7U chassis. Seeing that each chassis (using IBM here as an example) can pull 4kW of power, we could draw 24kW of power per cabinet. In most data center layouts, cabinets generally have one perforated tile in front of them which can supply 300 cubic feet per minute (CFM) based on std. design under-floor static pressure. To maintain an acceptable temperature rise through the server (so the exhaust isn't melting things), we need to supply approx. 120 CFM per kW of load. For 24kW that would be 2880 CFM. Seeing that the cabinet only has one 300 CFM tile in front of it, we have a problem Houston. There are variety of successful air-side strategies to deal with this including supplemental air delivery and exhaust air segregation. Ultimately, to go much higher than 24kW, we may need to use liquid cooling, which has the ability to remove far more heat, but with the liabilities mentioned by others. Cooling the room more doesn't help, as it's really an air delivery issue, regardless of supply temperature. In fact, overcooling is tremendously expensive due to the dehumidification effect of really cold cooling coils. The humidity then has to returned to the air to maintain the 50% RH goal. There are huge energy losses associated with this. Additionally, when the under-floor supply temperature goes below the dew point, some surfaces will as well, and they will have condensation occurring on them. Needless to say, condensation is not good for PC boards. In the short term, separation of the supply air and the exhaust air to avoid mixing is a good strategy. By avoiding mixing, assuming the available air volume is sufficient, high density power consumption can be cooled. A raised computer floor isn't even necessary for this if the air delivery method is designed right.