Modular UPS & Data Centre Virtualisation

Today’s business and economic climate exerts complex and dynamic pressures on data centres and their operators. Continuing growth in online e-commerce, social networks and mobile devices stimulates a correspondingly growing demand for resilient, always-available capacity. Yet this expansion is not steady or predictable. Economic factors randomly affect demand either positively or negatively, and loading is periodically also changed as new software packages are rolled out to deploy newer, more competitive or efficient features. Due to these unknowns, there is an inevitable tendency to provide further hardware to absorb unexpected peaks in demand. Purchasing and installing this additional hardware obviously increases both capital and operating expenditure, yet this pressure can be exacerbated as data centres are typically severely constrained for real estate. As a result, hardware density and cooling requirements increase – which both increase energy demand and therefore operating costs. Virtualisation Many data centres confronted with these issues have realised that these problems arise largely from underutilised hardware. Discussions in media, blogs and forums mention a utilisation rate of 10 – 15%, while rates down to 1% are not unknown. Such organisations have increasingly responded to this by using virtualisation. Virtualisation frees operating systems and applications from specific hardware. Instead, each operating system and application runs within a software container called a Virtual Machine (VM). VMs are isolated from one another, while hardware resources – CPUs, storage and networks – are pooled, and these resources are delivered dynamically to each VM by a software manager called a Hypervisor. Each VM receives the resources it needs for peak performance, even though several VMs may be running on a single hardware server. Hardware count and overhead is reduced, while applications’ performance improves. Virtualisation also allows data centres to offer higher availability and better fault tolerance. Organisations that adopt a virtualisation strategy can find that this has a somewhat novel effect on their UPS installation. Whereas they may previously have struggled to accommodate an increasing load, they now find that their UPS’s load has suddenly decreased as virtualisation efficiency allows hardware to be taken off line. This reduction in load can be significant, and if the UPS is a traditional transformer-based unit, its efficiency will be severely reduced. Such systems typically run at around 94% efficiency on a load of 80% or more, but drop to around 86% efficiency at 25% loading. This has severe impact on the data centre’s operating costs, both directly and because the increased heat losses associated with lower efficiency incur further expenditure for cooling energy. The data centre’s Power Usage Effectiveness (PUE) ratio will also be adversely affected, which can be a politically as well as commercially sensitive issue. Organisations today are increasingly expected to demonstrate a socially-responsible, ‘green’ image. Transformerless designs’ efficiency is far more load-independent, yet a successful virtualisation exercise will lead to excess UPS capacity on line unless action is taken. Meanwhile, the data centre will continue to face the other dynamic load factors mentioned above, as these will not be affected by virtualisation. Demand for capacity can continue to rise over time with more customers and more applications, so server hardware will start re-appearing on line to handle it, increasing the UPS load again. Over the long term, therefore, the UPS will experience a volatile load that can both decrease and increase periodically. A highly flexible UPS topology becomes essential to absorb these variations while minimising capital expenditure on UPS installation as well as operating expenses due to any loss of efficiency. Modular UPS Technology Fortunately, today’s UPS technology allows users to implement systems flexible enough to cope with these periodically decreasing and increasing load scenarios. For example highly modular solutions exist for installations of up to 1 MVA with N+1 redundancy. The elimination of the transformer and the 12-pulse rectifier employed on earlier systems means that a complete UPS unit can be built as a relatively small rack-mounting module. Each module can have a capacity as low as 10 kVA, with 20, 30, 40 and 50 kVA sizes also available. A single floor-standing UPS rack unit can accommodate up to five modules, so any capacity from 10 kVA to 250 kVA – or 200 kVA with N+1 redundancy - is possible. For larger systems, up to five such racks can be paralleled to deliver 1 MVA with N+1 redundancy. This modular approach has several dimensions of efficiency. Electrically, it offers 95% to 96% efficiency for any load down to 25%. Reactive power taken from the grid is minimised, with modular UPSs offering a near unity input power factor and a low input harmonic distortion of below 3%. Leading and lagging power factor loads can also be supported, as the modules are blade server friendly. The data centre’s precious real estate can be managed more efficiently, and future capital expenditure on excess UPS capacity can also be eliminated. These benefits arise because, once the decision to go modular has been taken, long-term forward planning is usually possible even when it is difficult to predict shorter term peaks and troughs. For example, if forecasting indicates that the data centre’s current 100 kVA load is highly unlikely to exceed 400 kVA throughout the centre’s operating life, then this scenario can be managed by purchasing, installing and cabling- in a couple of racks only. Alternatively, a single rack could be installed with space reserved for later installation of the second rack if needed. Footprint can be kept to a minimum, as a single rack with up to 250 kVA capacity occupies 0.43 m2, with a power density of up to 342 kW/m2. Initially, a couple of 50 kVA modules – or three for N+1 redundancy – could be installed into the first rack to run the existing load. As time goes by, more modules could be incrementally added to keep pace with increasing demand. Equally, modules could be removed to maintain UPS rightsizing following a virtualisation exercise and reduction in load. This virtually eliminates ongoing installation costs, because once the initial installation is complete and the racks are in place, modules are added in a plug ‘n’ play mode without need for cabling or building work. In fact the exercise is often referred to as ‘hot swapping’, because modules can be added or removed without even powering down the rack, in a procedure that can typically be completed within about half an hour. Modularity and hot-swapping also improves availability - another UPS benefit which is vital to data centre operators even if it does not concern efficiency. This is because availability is defined as MTBF/(MTBF+MTTR) where MTBF is Mean Time Between Failures, and MTTR is Mean Time To Repair. Reducing MTTR directly improves availability, and hot-swapping reduces MTTR to around half an hour as described. This compares with the approximately six hours required by monolithic systems which must be repaired in situ. As a result, modular systems can offer ‘six nines’ or 99.9999% availability. Conclusions As we have seen, a data centre is a dynamic and unpredictable environment in terms of its load conditions. Virtualisation can significantly decrease a data centre’s load on its UPS installation, while unpredictable changes in external requirements and internal software initiatives cause ongoing fluctuations in the load. Modular UPS topology has proven to be an invaluable solution to these conditions, allowing flexibility and accuracy in tracking these changing loads while minimising installation and operating costs, and maintaining the level of availability essential to today’s data centres.