Processing and power efficiencies are driving data centre developments. Kevin White investigates.
IT power costs have been rising steadily by 15% a year for the past five years and forecasters estimate they will exceed server procurement costs within the next five years, unless something is done.
IT executives know that something has to give but many are caught having to tread a fine line to balance their budgets, needing to sweat their server assets for as long as possible while wanting to swap them out for newer, more energy-efficient lines.
“As IT and facilities budgets receive increased scrutiny, efficiency becomes a more business critical issue for data centre managers globally,” confirms Tom Brey, a director of The Green Grid. Formed in 2007, this is one of several cross-industry bodies that is lobbying to advance energy efficiency in data centres for business computing.
Led by systems vendors, but also sponsored by businesses as diverse as the Nationwide US insurance company, Walt Disney and eBay, the aim of Green Grid is to develop a top-to-bottom blueprint for energy-efficient data centres.
One of the architectural building blocks of the new data centre are low-cost commodity servers based on the Intel Xeon Processor 7400 Series, which can be easily and cheaply configured as high-performance platforms for virtualisation.
Lines like the IBM System x3850 M2 have a lot going for them, designed as they are to support large numbers of virtual machines capable of running multiple business applications on a single server.
System x3850 M2 comes VM-ready: the servers can be deployed with VMware ESXi embedded to provide virtualisation capabilities right out of the box, with 40% better virtualisation performance than previous quad-core generations. And importantly, this is within the same power envelope.
Benchmarked against comparable Proliant 580 and HP Blade 685 machines, the line has been judged to offer high consolidation ratios and will scale to a ratio of 227:1 in a virtual desktop infrastructure scenario, allowing the workload of over 200 PCs to be collapsed onto a single box.
Willy Chiu, VP for High-Performance On-Demand Solutions at IBM, believes the industry is now very well placed to be able to draw on its 40 years of experience with virtualisation, and entirely transform enterprise compute processes.
“It’s important for any organisation to be able to marshal all the technology resources available to it at any one time. The goal is virtualisation across all platforms, and a dynamic infrastructure that will eventually represent the next-generation data centre with an entirely different footprint from today’s,” he said.
IBM has around 8 million square foot of data centre resources to draw on around the globe, and Chiu describes how the company is working to be able to string them all together through virtualisation and cloud-based provisioning so as to be able to dynamically manage a server workload to wherever it can be most efficiently processed. “Ultimately, it’s all about changing the economics of enterprise computing. It’s all about boosting efficiency. And it’s about doing more with less.”
System vendors are in agreement that there are many different classes of workload that can easily be virtualised to take advantage of new data server technologies like System x3850 M2, and the virtualisation and cloud-based approaches these and other boxes underpin.
Chui takes up that theme: “We estimate that in a typical enterprise around 40% of server resource can be tied up in test and development processes. That’s one very good example of a workload that could be virtualised and spread across different servers. We estimate that around 80% of the enterprise application estate is movable in that way.”
Studies show organisations that work towards optimising their data centre environments using virtualisation can quickly realise benefits. On average, they can expect a 4:1 reduction in the number of servers needed to handle their workloads, and a 7:1 decrease in the cost of administrative support needed to run them.
Consolidating services onto fewer physical servers undoubtedly helps compress running costs and power requirements. Not only is the cost of power rising relative to the cost of computer hardware and software, but its future availability is far from certain in some circumstances.
Local power grids might not always able to cope with the appetite of large server farms and data centre operations in major cities such as London. The national grid in south east England is actually near to reaching breaking point. Electricity supplier EDF has already warned businesses in the region that it is unlikely to be able to increase supply to data centres at short notice.
The power supplier now requires formal notification of any new data centre or expansion of existing data centres.
HSBC is one big name company that is prepared to invest in a green future, having publicly stated an intention to become the world’s first carbon-neutral bank. One of the main ways the group is working to reduce its carbon footprint is through the use of technology to reduce unnecessary travel. But it is currently also looking at new ways of constructing data centres in Mexico, Hong Kong, the UK and US that will be much leaner, energy-efficient power houses for its global business.
Servers can consume energy at an alarming rate. A single server may use as much electricity as six domestic fridges and a medium-sized data centre of 1,000 servers may be responsible for 3,500 tonnes of CO2 per year. Suppliers are doing their bit, and Computacenter was one of the first to say it would bring the Electronic Product Environmental Assessment Tool (EPEAT) system for identifying environmentally friendly IT kit to the UK.
Designed to make it easier for IT shops to measure the energy footprint of their IT estate, EPEAT is used to rank desktops, notebooks and monitors on a range of environmental criteria but perversely there is no equivalent system for bigger more power-hungry servers.
It has been estimated that up until now, for every pound spent on a server there will be another 52 pence spent on powering and cooling it. Bodies like Green Grid are currently working on new metrics for estimating data centre productivity, in the realisation that the heating and ventilation demands of servers need to be accurately measured before they can be managed down.
For its part, IBM is looking at different ways of optimising the energy and thermal environment in the data centre for cost and performance efficiency. Tivoli Monitoring for Energy Management is one development being proposed that will let system administrators visualise the power consumption and thermal signatures of their data centre resources.
The introduction of IBM Tivoli Monitoring Green Energy adapters will mean IT managers could also start to understand the business service impact of their energy optimisation decisions, and pull energy usage information into Tivoli to enable energy-related consumption and chargeback schemes.
Knowing how much energy is being used is an essential starting point in the process of optimising it. From there a benchmark can be set for energy usage, so that improvements can be better tracked.
Dave Mitchell, an analyst responsible for Ovum’s global research activities, sees the proposed extension of the system management role into the province of data centre HVAC (heating, ventilation and air conditioning) control as one that holds lots of promise. “Ultimately, we want to be moving to a position where we are provisioning application workloads to systems based on the power efficiency of the supporting servers.”
He explained that the power efficiency of servers is anything but linear, and that tighter systems management controls would be one way of better optimising server use for energy efficiency. “There seems to be a sweet spot for transaction servers when the I/O is running at 50% and CPU utilisation is around 70%,” Mitchell states, suggesting that every data centre administrator would benefit from having an ‘optimise for power’ button on their console, something that might automatically handle workload distribution for power efficiency.
Lines like the x3850 have been designed to be more power-efficient than comparable rack or blade systems. Most of the power lost in a data centre occurs during conversion of the AC to DC supply, with as much as 35% to 40% of the energy going into the system lost through heat. IBM reckons the x3850 is 92% efficient in terms of its power supply.
Another development that has come out of the IBM Labs is a system of mobile measurement sensors that provide real-time and high resolution 3D thermal and energy maps of a data centre. In tests it has repeatedly been shown that data centre energy savings of more than 10% are possible through some simple readjustment of racks and targeting hot spots with cooling snorkels and other low cost ventilation options. This too, will integrate with Tivoli for automated environment control, data centre floor planning and energy management.
Energy consumption and processing efficiency will play an increasingly important role in server refresh decisions, and latest generation servers use processor, memory and power conversion supply systems that will drastically cut electricity bills. Green IT is good for the bottom line and is good for the environment, and the power savings that stem from choosing power-efficient servers are unlike other total cost-of-ownership criteria in that the savings are immediate.