Optimization of cluster cooling performance for data centers

Abstract

A software tool using a Genetic Algorithm (GA) has been developed to optimize the cooling performance of a cluster of equipment comprised of two approximately-equal-length rows of racks and coolers bounding a common hot-aisle. Such clusters are "room neutral" from a cooling-load perspective if most or all of the hot rack exhaust is captured locally by the coolers. A direct cooling-performance assessment relative to this design objective is provided on a rack-level basis by the Capture Index (CI) and on a room-level basis by the Total Escaped Power (TEP). Various prediction techniques can be used to compute these cooling performance metrics for a cluster in real time. In this study, a Neural-Network-based algorithm is used as the "cooling prediction engine" upon which the GA optimization functionality is built. Typically, optimization of cluster cooling performance is attempted by intuition or trial and error, and one has to rely on the designer's experience. In contrast, the GA-based optimization technique in combination with the NN calculator provides a systematic approach for finding an optimized cluster layout. This integrated tool can be used to find the best arrangement of a fixed inventory of racks and coolers. This approach can also be used to find the best distribution of total heat load among racks and to identify the best rack location to which extra heat load can be added. Examples are discussed to demonstrate the potential of the software tool.

Multi-objective efficiency enhancement using workload spreading in an operational data center

“…Machine learning based models have been explored for prediction of air temperature at discrete points, such as inlets and outlets of servers for a new workload spreading method [16][17][18][19][20]. In this method, a heat flow model of the DC is created based on experimental observations made inside the DC in order to correlate a relation between the workload of servers and their inlet and outlet temperature.…”

Genetic algorithm based cooling energy optimization of data centers

“…Studies that simultaneously optimize thermal management and cooling energy consumption can be classified into two categories, dynamic optimization and static optimization, depending on whether the optimization framework runs in (near-)real-time. Studies using static optimization involve optimizing data center layout (Shrivastava et al , 2008; Li et al , 2007), as well as determining suitable ranges for data center operation parameters (e.g. target temperature rise across servers and computer room air conditioning (CRAC) units) (Shah et al , 2004a, 2004b, 2005a, 2005b, 2008).…”

“…This rack-level optimization study only considered thermal management; minimization of cooling cost was not considered. Optimization of room cluster layout based on parameters such as rack capture index (CI) and room total escaped power (TEP) was considered by Shrivastava et al (2008). CI is defined as the fraction of hot air exhausted by a rack captured by return vents, while TEP is the fraction of heat dissipated by the entire cluster that is captured by in-row coolers.…”