Scalable Data Centre Efficiency

Session 13, Paper 3, CIBSE Technical Symposium, Liverpool John Moores University, Liverpool
11-12 April 2013

The data centre industry has been improving its energy efficiency in recent years. Operators compete to differentiate their performance, often using the PUE metric to demonstrate their ‘green’ credentials. Data centres differ from other building types in that they are not usually fully loaded immediately after completion and handover. The IT load grows over time and may take several years to reach design load (if ever), with the facility operating for considerable periods at part load. It is therefore important for designers to consider how their infrastructure scales with load. Any design will have elements of fixed and variable overheads for power and cooling systems; the best scalability is achieved when the fixed losses are minimised. These fixed losses result in a worse PUE at part loads (it tends to improve as load increases), although part load in itself need not be a reason for poor performance. By analysing the energy breakdown of a facility at its present load condition, it is possible to benchmark performance, independently of PUE and identify facilities with high potential for efficiency improvements. This allows facilities with different designs, ages and actual performance to be compared, for example, within an operator’s portfolio, although their starting points and energy improvement solutions may be different. Using these techniques, operators can focus their energy optimisation activities.