The Power to be Cool
Evolving the Data Centre for a new set of challenges
Anyone working in the ICT sector these days will be well aware of the constant exhortations to improve our green credentials. Indeed, if you've been reading the press you'll find out that Logicalis too has been adding its voice to the growing debate.
Environmental concerns, along with growing legislation and new competitive pressures, are just some of the factors that are focusing attention on the rapidly growing costs of operating complex Data Centres. This, in turn, is putting mounting pressure on ICT departments to convince their businesses that they are utilising resources to the maximum, yet still able to provide an assured service across the entire suite of information and communications applications.
Over many years, the number of applications and services the Data Centre delivers has continued to increase. And, with this growth, the original Data Centre, designed for a single mainframe or a few large servers and a few tape drives, has morphed into a complex connected multi-disciplined beast.
Growing power consumption and cost
Against this background, Data Centres are consuming resources at an ever growing rate. In fact, a third to a half of all corporate energy consumption in the UK is now down to IT systems and their infrastructure.
With energy costs rising at 30 per cent each year for the last three years, just providing electricity to a large Data Centre has risen to an estimated 50 per cent of its total costs. Against this background, the assertion by The Broad Group, that suggests, that the power costs for an average Data Centre will double from £5million in 2006 to over £10million by the end of the decade looks positively reassuring.
And the ICT industry's dirty little secret is starting to leak. According to Forrester Research a Data Centre with 25,000 servers will use enough electricity in a single month to power 420,000 homes for a year. Even a Data Centre half or a quarter of this size is clearly eating up huge amounts of power and money and these sorts of comparisons are being noticed by those driving legislation. Expect Ken Livingstone to turn his attention to Data Centres as he attempts to deliver on his targets for carbon footprint reduction in London . Expect too changes in corporate accounting. Soon CIOs will be in charge of paying their own electricity bills!
An upward spiral
One of the reasons for the growth in electricity consumption is that Data Centre operators have been fighting to find the space to meet burgeoning demand for processing capacity, by using high density computing approaches based on compact blade server technology. The effect of this is that Data Centres originally specified to supply a maximum of 1 kilowatt per square metre are now consuming as much as 10-15 kilowatts in the same space. This, of course, is driven by growing demand for processing and storage but dense packing has multiplied both power and cooling requirements - resulting in an upward spiral of electricity usage.
Indeed, as chip performance shows no signs of disobeying Moore 's law, the cost of cooling a server over its life time is now greater than the initial cost of the technology itself. Now, the conventional wisdom is that for every watt used to power a device another watt is used to cool it. And the cost of cooling is still climbing relative to power.
Dealing with the dilemma
Such is the increase in power consumption that Gartner has predicted that half of the Data Centres in the world are going to run out of power by the end of 2008. Even in our 24/7 economy that is just 245 days! But, of course, it would be unrealistic to expect organisations to slow the march of ICT within their organisations. Instead we need to turn attention to the technologies that can significantly reduce electricity consumption. Ideally, these should be housed in brand new Data Centres built from the ground up, to accommodate the demands of modern ICT capacity. But the fact is that Data Centres require resources that are in very short supply in the UK , particularly in the South East of England where many businesses are based. Cheap real estate, access to power, let alone cheap power and communications bandwidth are all key factors preventing the building of newer, bigger, better more energy efficient Data Centres.
As a result, we are faced with using existing Data Centres, where wholesale replacement is not an option, but where existing systems and space are so stretched that many are operating a policy of 'no equipment in without old equipment out`. We have to focus firmly on increasing efficiency and the rapid evolution of the facilities that are already available.
Sheer power versus efficiency
Moore 's law has driven us down a track of absolute power - essentially more processing capacity from more electricity consumption. Hence the historical ability to rapidly increase the MIPS per square foot ratio with all its impact on cooling. What we now need is efficiency - more MIPS from fewer watts and more availability from fewer devices. But how is this achieved? There are three basic but interconnected routes. Less power-hungry technology; better cooling and more efficient use of resources.
Intel and AMD, for instance, have focused on on-device power management in their 64-bit development programmes, with IBM looking to silicon-on-insulator technology to reduce wasteful leakage from circuits. Other suppliers such as Sun have used multi-threading technology to finely tune instruction execution in heavy transaction processing environments, vastly increasing the MIPS/Watt performance and reducing the need for as many processors.
Really not cool
Older Data Centres deliver cooling in a way that has changed little since mainframes dominated the environment, with air circulating in a relatively unmanaged way, being cooled at floor level only for it to be heated by hot server racks again before it reaches the exit vents. By modern standards this is not in itself efficient. But with the move to rack-mounted blade-based server systems, which are constantly being repurposed or repositioned to keep up with the demands of business, or to maximise existing space in the Data Centre, such a system fails to cope. A much better job needs to be done of monitoring, analysing and dynamically managing airflow in discrete enclosures.
New generations of air conditioning systems already on the market use sophisticated matrices of sensors and fans that can work in conjunction with new rack infrastructures, such as the air-tight blade chassis now available from Logicalis partners IBM and HP. HP's C-class series, for instance, optimises the efficiency of air cooling by managing air flow inside as well as outside chassis enclosures. With the addition of complementary power management technology HP believes its C7000 chassis could cut as much as 60 per cent from operating costs, through power use reduction. Meanwhile, with Blade Centre, IBM has reintroduced water cooling to the Data Centre taking systems that exist in nature as its inspiration for new, more efficient ways of heat dissipation.
Physical and virtual
These sorts of direct, physical approaches will continue to impact the issue of runaway power consumption, but these methods may require substantial capital expenditure and may have to be tied to replacement cycles. A fast and cost-effective approach to the problem is simply to reduce the number of devices that are contained in the Data Centre. And the key to this is logical virtualisation.
Virtualisation enables multiple instances of one operating system to run on one physical machine, or alternatively by making it appear is if multiple physical resources such as disks or memory are one virtual device. This has been applied to high end mainframe and RISC servers for some years but has recently been made available so users can rationalise their Intel-based server estate and storage devices.
Virtualisation breaks one of the conventions of ICT - that of over-provisioning. This amounts to a wasteful constant resourcing of servers and storage to ensure that they can cope with infrequent, sometimes unpredictable spikes in the workload. Like leaving your car idling all the time, just in case you want to use it.
This approach has the potential to cause massive waste. If a system is running at 15 per cent of its capacity - which is very common - it may still be consuming 50 per cent of the power it would be using at peak utilisation. In a virtualised environment multiple applications would share a common set of resources accessing as much or little of the total capacity as required.
The improvements can be dramatic. Managed properly, virtualised servers can be run at 80 per cent of their peak capacity resulting in a drastic reduction of idle time. And there is an important multiplier effect at work. The need for fewer servers inevitably means less power consumed running the devices themselves, in cooling them and in powering other support functions such as humidification and power protection systems. And it releases physical space for alternative technologies or cost reduction.
This can make a very rapid impact on the current situation where less than half the electricity going into a Data Centre is servicing an ICT workload. Virtualisation too need not stop in the Data Centre, of course. Virtualisation of the desktop alone could save £70 in electricity costs a year, just by preventing users leaving their PCs on overnight.
It's a Data Centre, Jim...
So, the Data Centre should be firmly back at the centre of the corporate ICT agenda. But to paraphrase Star Trek's Spock 'it's a Data Centre, Jim, but not as we know it`. A rapid evolution of its capabilities and efficient management of the resource is essential if it is to support the organisation - both enabling emerging ICT environments, and contributing to overall efficiency, compliance and environmental performance of the organisation as a whole.
5Back to Top 5Back to Homepage
