15 March 2007
As the normal running load of power systems rarely exceeds 40 per cent, the importance of efficiency at partial-load is becoming more important than the full-load efficiency in critical data-centre applications, as Ian Bitterlin’s technical paper explains
THIS PAPER REVIEWS THE LOSSES IN A CRITICAL POWER SYSTEM for competing UPS topologies -double conversion static UPS supported by lead-acid batteries v line-interactive static UPS supported by medium-speed flywheel energy storage.
This article reviews a Tier IV power system – comprising, as it should, dual-bus power architecture with N+1 redundancy in each critical power path. The system proposed is a true A+B dual bus power solution. Each path is active (in normal operation providing half of the power demand) and each is capable of supplying the full load. It is a ‘2(N+1)’ solution by definition. The A&B power systems should be compartmentalised and the power distribution routes should be physically diverse.
For our comparison, the system capacity is matched to the size of the critical space (425m2) and the design power density (3kW/m2) and is fed by two 1,275kW UPS systems. It is strongly recommended to view IT loads in kW rather than kVA since the loads are tending towards unity Power Factor and even slightly leading. With UPS’s historically being rated at 0.8PF this corresponds to 1600kVA. Each system provides 100 per cent redundancy for the other and each has at least N+1 redundancy.
In new systems the IT load will develop with four characteristic variables:
● Growth in capacity, e.g. from Day 1 at around 20 to 80 per cent after 12-18 months
● Balance between single power supply and dual-corded devices
● Technology changes in IT hardware that may affect the growth in power, and
● Electrical characteristics, e.g. harmonic current content, crest factor and (displacement) power factor.
If the single corded loads are well managed and distributed evenly across the dual-bus system then each system will commence duty at 10 per cent load and rise (at one rate or another) to around 40 per cent load when the zone is fitted out.
The management of the single-corded loads can be important. Consider the case where the proportion of single-cord load is 30 per cent of the total and it is poorly managed to 1/3rd on bus A and 2/3rd on bus B. Under these circumstances the system loads will be 36 per cent and 44 per cent respectively. It is easy for the imbalance of single-corded loads to create a scenario where the load step on transfer is greater than 50 per cent and this should be avoided.
If you take the annual maintenance shutdown as the only significant power event and it lasts four hours, then for 99.95 per cent of the life of the UPS systems they run at 40 per cent load, or much less if not managed and fully populated. An N+1 modular system that can grow or decline in capacity to follow the load profile has many advantages in this scenario.
The system load is carried by the UPS modules dependent upon the integer of the number N in the N+1 architecture. The table (below left) shows the results, including the (risky) 100 per cent load condition. It is immediately apparent that the operational performance, and particularly partial load efficiency, of the individual UPS modules at loads between 25-40 per cent is a critical for the running costs and, as a consequence, energy efficiency and resultant carbon emissions.
In general the higher the UPS module load the higher is its efficiency and so, from an energy efficiency standpoint, the higher the number of modules (each at a higher load percentage) the better. The opposite is true of the theoretical reliability since the component count increases, although with dual-bus and N+1 in each bus this is of low consequence. Also it is worth noting that the higher the number of modules the lower is the maximum step-load that each is required to take when the entire load transfers to one system in one step.
We can compare average industry efficiency performance data (at 25-100 per cent load) for various static UPS topologies and technologies that present relatively low harmonic distortion to the mains supply:
● Transformer-less series on-line double conversion with IGBT rectifier and lead-acid battery energy storage
● Series on-line double conversion with thyristor rectifier, 5th harmonic filter, transformer and lead-acid battery energy storage
● Line-interactive topology with medium speed flywheel energy storage
These data points can be better viewed in graphical form, (above), and for a simple comparison of losses we can take the average efficiency of each topology at 40 per cent module load and convert to kWh/yr. The difference in power losses (including the power used to remove the heat from the UPS room and reject it to atmosphere) between the IGBT rectifier model and the line-interactive flywheel topology is circa 500,000kWh/yr, equivalent to €45,000 at a nominal cost of €0.06/kWh – or equivalent to 47,000kg of CO2 emissions per year if burning coal to generate the power.
It is worth noting that the partial-load efficiency of most forms of rotary UPS, both battery and diesel supported, have a lower efficiency at 25-50 per cent than those shown above. The difference in losses can result in a payback time of as little as four years depending upon the local cost of electricity and the market price of UPS hardware.
The limiting factor on the size of many installations is the available capacity of the grid supply and the associated transformer. A 5 per cent improvement in the efficiency of the UPS then brings opportunities to save further costs or to improve the business model:
● There is 5 per cent more usable power for IT power or cabinet space
● There is 5 per cent more usable power for cooling energy if the load density is high and cooling plant such as fans or pumps need UPS power
● Higher efficiency means lower input current so that cable or busduct cross-sections can be optimized.
It is clear that energy losses are coming under increasing pressure from government legislation around the world. The need for carbon emission reductions to mitigate the acceleration of global warming is well proven. Datacentres, by their very nature, are huge consumers of power and rejecters of waste heat. In classical engineering terms their efficiency is less than zero, since for every 1.0 Watt that a microprocessor consumes the data-center will consume and reject to atmosphere between 1.6-2.0 Watts. With not atypical facilities being built with IT loads of 10MW (and substantially greater) any contribution to saving energy from the power or heatprocessing infrastructure will become increasingly important. Both the monetary cost and the environmental impact will interject into the traditional discussions on reliability and availability.
The latest UPS mainstream technologies (IGBT rectifier) and topology (transformer-less series on-line) fall short of what is possible and what is desirable in a Tier IV class critical facility. Unlike many ‘green’ issues these days this environmentally friendly option actually is the lower cost option in the short to medium term, rather than the long term to ‘never’ unless government subsidies apply.
The Uptime Institute has, for more than 10 years, sponsored research and practical studies into data centre design, operation and resultant resilience and developed a Tier Classification to describe and differentiate facilities from an Availability standpoint. A White Paper from the Institute forms the background of this paper.
1 The Uptime Institute, Building 100, 2904 Rodeo Park Drive East, Santa Fe, NM 87505, USA
2 Title: Industry Standard Tier Classifications Define Site Infrastructure Performance, Turner, Seader & Brill, © 2001- 2005 The Uptime Institute, Inc