CFMs, cubic feet per minute, for a data center is the amount of air being delivered to the cold aisles. CFMs are also discussed as the amount of air produced by fans internal to components of a HPC system. If the CFMs being produced by the system exceed the maximum CFMs a data center can deliver to the cold aisles, hot air may be circulating into the cold aisles, can create hot spots in a rack or in certain sections of the data center, increase the likely hood of component failures, and possibly exceed the temperature limits of components causing power outage and downtime.
One key formula I use when I want to calculate CFMs for particular components of a HPC system, such as a compute node, is the following:
CFMs = (3412*W/1000) / (1.085 * ∆T)
In the above formula, W represents the peak measured power consumption in watts when stressing the components with a particular benchmark, such as Linpack discussed in Part 2 of my blog, and ∆T is the change in temperature between the air entering the system components and air existing the system components. When the change in temperature is large, typically seen when the inlet temperature is low, the CFMs are lower. But if the change in temperature is small, typically seen if the inlet temperature is high, than the resulting CFMs is higher because the fan speed increases to remove the hot air quickly from the components. Typical measured change of temperature I personally seen at customer sites is anywhere from 20F to 30F, and can be as high as 40F to 50F, but this depends on each customer’s data center, the inlet temperature, and how well the hot aisle is contained to avoid hot air leaks into the cold aisle.
Another general rule of thumb I also have used, and has worked well, is 100 CFMs for every 1 kilowatt of power measured.
In part 5 of my blog, the next topic to discuss for a HPC system is British Thermal Units per hour (BTUs/h), which is a measurement of heat produced.