A Rack-Level Cooling Redundancy Metric for Data Center Applications

Abstract

Redundancy is an important measure of an operation’s ability to withstand planned or unplanned system failures. While this concept is commonly used in power systems, redundancy can be extended to data center cooling systems, as well. We propose a rack-based redundancy metric for cooling performance that is similar in nomenclature to metrics for power systems, but also captures the local nature of data center cooling. This paper will explain how to compute this metric for general data center layouts and show how cooling redundancy can influence design choices when used in combination with typical measures of cooling coverage: inlet temperature and Capture Index.

Failure Resistant Data Center Cooling Control Through Model-Based Thermal Zone Mapping

“…In order to ensure "N+k" cooling influence redundancy, the third thermal zone partitioning criterion in the previous subsection can be employed with m = k + 1, which ensures each rack inlet temperature is covered by at least k + 1 thermal zones, and hence even in the case of k failing CRAC units there is still at least one functioning CRAC unit trying to maintain each specific rack inlet temperature at desirable thermal status. Note that "N+k" cooling influence redundancy only ensures that each rack inlet temperature is continuously monitored and controlled by "k+1" CRAC units, but does not necessarily lead to "N+k" cooling redundancy as defined in [2] that guarantees satisfactory data center thermal status in the event of "k" failed CRAC units in any possible combination. Cooling influence redundancy focuses on using available cooling resource for improved cooling redundancy, and hence does not have any specific requirements on the size of cooling resource pool.…”

“…ASHRAE [1] defines a room level cooling redundancy of "N" as a cooling system that sufficiently cools a room, and "N+k" cooling redundancy as the addition of extra "k" cooling units. This definition, as pointed out in [2], does not differentiate between important cooling system configuration parameters such as size, cooling capacity, and placement of the cooling units. Recognizing these drawbacks, rack level cooling redundancy is defined in [2] as the number of simultaneous cooling unit failures in all possible combinations that a particular rack can withstand.…”

“…This definition, as pointed out in [2], does not differentiate between important cooling system configuration parameters such as size, cooling capacity, and placement of the cooling units. Recognizing these drawbacks, rack level cooling redundancy is defined in [2] as the number of simultaneous cooling unit failures in all possible combinations that a particular rack can withstand. Although a big leap from the previous ASHRAE definition, the cooling redundancy definition in [2] does not clearly state whether the cooling units have their respective fixed cooling resource provisioning setting or whether they are allowed to change the cooling settings and adapt to the current thermal status.…”

“…Recognizing these drawbacks, rack level cooling redundancy is defined in [2] as the number of simultaneous cooling unit failures in all possible combinations that a particular rack can withstand. Although a big leap from the previous ASHRAE definition, the cooling redundancy definition in [2] does not clearly state whether the cooling units have their respective fixed cooling resource provisioning setting or whether they are allowed to change the cooling settings and adapt to the current thermal status. This question leads to the distinction between what the authors of the present paper call "static redundancy" and "dynamic redundancy".…”