Server Data Optimizes Cooling
November 27, 2012, Server Design Summit, Santa Clara, CA—Akhil Docca from Future Facilities presented concepts for using data from the servers to control the cooling for the systems. Some of these data are already used for monitoring and control, so this is mostly a redirection of data from other purposes to cooling.
Current chips have the capabilities of self-monitoring, and the necessary metrics for system performance optimization are already available. The tools for optimizing space, power, cooling, and networks are in place, but the actual or useable capacity does not match the intent of the designed-in functions. Part of the difference is due to changes in plans, operations, and hardware before the system is completely refurbished.
The problem with most tools for the data center is that they don’t account for the mix of used, available, and lost capacity. Space and power use is determined by various limits like power per slot, cooling per rack, and network capacity per shelf. The piecemeal replacement and upgrade of servers causes hardware and software fragmentation, and the loss of capacity. Space and power are insidious, because the other constraints are considered to be integrated over time, but are actually discrete steps.
Even worse, cooling is invisible. The IT industry uses air cooling as the baseline for installations. Each piece of equipment is allotted some amount of airflow at a temperature. One issue is that there are no standards for the cooling, some units force air in, others force air out, and exhausts are on top, side, or back of a unit. There is no easy way to match the actual airflow to the equipment airflow requirements, and misdirected exhausts may be feeding an intake port.
The recirculation or bypass for the various airflow options is hard to model and even harder to control. There are no links between servers and the data center facilities for cooling control. What is needed is an integrated thermal design capability that can address the problems of data center thermal design. Such a tool has to encompass computational thermal models and actual temperature data from the servers to improve overall cooling and reduce the need for large steps in cooling capacity.
Some of the necessary functions for the tools include data aggregation of data, predictive analytics, reports and monitoring, policies, and troubleshooting. These can all be integrated into a telemetry platform with fault control for better system management.
For example, a thermodynamic analysis would enable visualization and virtual prototypes to configure a new installation. The ability to see the airflow and the results of changes in that flow can result in a dynamic system that changes airflow and air temperature with demand. The result is the potential to increase the air temperature by up to 8°C and get a significant reduction is cooling costs. Another variant is to bring in cooler outside air at night for cooling.
