Networks Europe Nov-Dec 2016 | Page 36

losses will exist. As the IT load shrinks (e.g., from consolidation) these fixed losses become a higher proportion of the total data center energy use. This means PUE will worsen. This ENERGY also36 means that PUE is alwaysMANAGEMENT better at higher IT loads and worse at lower loads. Figure 3 shows a typical PUE curve illustrating the relationship between efficiency and the IT load. 5 Worse PUE 4 Efficiency degrades dramatically at low loads due to “fixed losses” inherent in unused power/cooling capacity 3 2 Better 1 0% No load 20% 40% 60% IT Load 80% 100% Full load % of data center’s power capacity Typical data centre infrastructure efficiency curve infrastructure to take account of a reduced IT load if they are to maintain a satisfactory PUE result. Consider by way of example a 1MW data centre with 1000 physical servers, each of which draws 250W of power. If these were to be virtualised with a conservative consolidation ratio of 10:1, with each physical server operating at a CPU utilisation of 60% (instead of the typical 5-10% for a standalone server) the overall energy savings over the course of a year could be 75% If the power and cooling infrastructure is left unchanged, PUE will increase after virtualisation. Adjusting these values, the top part of the PUE ratio, is a challenge because there are inherent fixed losses in power and cooling equipment that can’t be reduced linearly with load; they will be incurred no matter what the load is. Therefore as IT load shrinks, thanks to virtualised consolidation, these fixed losses account for a higher proportion of the total energy consumption of the data centre. The variation of PUE with the IT load of a data centre is represented as a curve with PUE degrading dramatically at lower IT load utilisations. To improve PUE after virtualisation it’s necessary to address the issue of fixed losses. This means that as far as possible, power and cooling capacity must be scaled down to match the load. This can be best achieved in a new data centre by following a standardised and modular approach to designing the infrastructure; making it easier to switch off unneeded cooling units, slowing down or turning off fans and deploying scalable UPS systems. Careful design of the cooling architecture, including containment alongside the use of hot or cool aisles, helps to manage capacity better resulting in a reduction in losses. Existing data centres For existing data centres, such a modular approach may not be possible. However there are other options to be considered Rev 5to IT to make the cooling effort more scalable with regard load. Blanking panels and air containment solutions could be installed to reduce or virtually eliminate hot and cold air mixing to improve the efficiency of cooling equipment. Adjustable fans could be installed to better match cooling effort to IT load. Unused power modules in UPSs should be removed. With a reduced IT load, it may also be possible to turn off one or more CRAC/CRAH units to further reduce infrastructure losses. It should be reiterated that PUE and overall power consumption are not necessarily contradictory concepts; the misconception can arise only when PUE is thought of as a green metric in itself, which it’s not. PUE takes no account, for example, of how the electricity for a data centre is generated – or how much is generated. It’s irrelevant (to PUE) whether power comes from a green renewable source such as solar or wind power or more traditional environmentally damaging sources such as oil or coal. PUE is merely a measure of how efficient power and cooling systems are for a given load. Its importance is only increasing, thanks to growing awareness among industry and the public at large of the need for greater energy efficiency. For truly environmentally friendly data centres, which also operate cost effectively and reliably, PUE and overall consumption have to be considered as separate, though inextricably related topics, and designers should optimise their facilities for both. n Schneider Electric – Data Center Science Center White Paper 118 5 www.networkseuropemagazine.com