Conditional Probabilistic Airflow Prediction for Data Center Cooling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional data center cooling systems are inefficient due to over-provisioning for peak load scenarios, leading to excessive cooling and high operational costs, as they do not account for the actual usage levels of computer systems, resulting in unnecessary energy consumption.
Innovation Solution
The implementation of conditional probabilistic models, such as Bayesian networks, to predict airflow rates through air delivery devices based on the airflow rates of air moving devices, allowing for optimized airflow distribution and minimizing the need for extensive sensor networks, thereby reducing costs and improving efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If cooling is provisioned for worst-case or peak load scenarios, then cooling reliability is improved, but energy consumption increases
Solution Approach 1:
The cooling system dynamically adjusts airflow rates based on real-time measurements of actual IT equipment power consumption. The system transitions from static peak-load provisioning to dynamic adjustment, where cooling capacity matches actual thermal loads, thereby maintaining reliability while reducing energy waste during partial-load operations
Solution Approach 2:
The system implements feedback control by continuously monitoring IT equipment power consumption and using this information to adjust cooling airflow rates. This closed-loop approach ensures cooling reliability is maintained while optimizing energy consumption based on actual thermal demands rather than worst-case assumptions
2Measurement precision
If extensive sensor networks are deployed to monitor airflow rates, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The system uses power consumption measurements as an intermediary variable to infer airflow requirements. Instead of directly measuring airflow with complex sensor networks, the system measures electrical power consumption of IT equipment and uses this information to determine appropriate cooling airflow rates, thereby reducing sensor complexity while maintaining measurement precision
Solution Approach 2:
The system replaces mechanical airflow sensors with electrical power measurement instruments. By substituting direct physical airflow measurement with electrical measurement of power consumption, the system achieves equivalent information about thermal loads with simpler, more reliable sensing infrastructure
Data Source
AI summary
In a method for predicting an airflow rate of at least one air delivery device, a plurality of airflow rates through the at least one air delivery device at a plurality of airflow rates of at least one air moving device are received. A conditional probabilistic model is generated with the air moving device flow rates as inputs and the airflow rates through the at least one air delivery device as outputs. Moreover, the airflow rate of the at least one air delivery device is predicted from the conditional probabilistic model.