Energy-efficient datacentre technology

The rising cost of electricity, combined with current or anticipated supply constraints and legislative pressure to reduce the carbon footprint of the IT estate, is forcing corporates and service providers to find ways to cut the amount of power their datacentres use.

New builds can be fitted with the latest technologies and design methodologies to ease the problem. Social networking giant Facebook is to spend $450m (£290m) on a new custom-built datacentre in Salt Lake City, for example, which it claims will use innovative cooling and power management technologies – though the company has not yet said what they will be. But upgrades to existing datacentre facilities can be trickier to implement, particularly given the budgetary pressures that many organisations are under.

Innovative technologies offering a quick route to hardware consolidation, such as server virtualisation, have already been widely deployed, though low physical server utilisation rates indicate there is still some way to go (research firm Gartner estimates that x86 servers are running at only 12 per cent utilisation, with racks populated to only 50 to 60 per cent capacity). But there are lots more steps that datacentre managers can take.

One such step is storage virtualisation, where hard disks from multiple physical Raid arrays or systems grouped together under a single virtual or logical entity, either file or block level, can be accessed irrespective of their location on the storage area network.

This approach improves hard disk utilisation, with additional portions of disk made available to create a new virtual disk, for example, and allows data to be migrated from one system to another without interrupting input/output (I/O) access to the disk.

Storage virtualisation can be supplemented with consolidation technologies such as data de-duplication, which minimises the volume of data to be stored and backed up to disk by removing multiple copies, as well as thin provisioning, which optimises available storage space by allocating blocks of capacity only when an application actually calls for it, rather than reserving larger chunks for future usage.

Server hardware: pack it in
Where appropriate, datacentre managers can also replace existing rack and blade servers with up-to-date, more energy-efficient hardware. Intel and AMD are constantly revamping server CPUs, chipsets and other system board components to use less power and space. Successive generations of server architecture have seen consistent improvements in energy efficiency – often by shrinking overall server size so that more can be packed into a single rack that can make better use of more efficient cooling technologies.

While this type of upgrade is unlikely to reduce electricity consumption per server, it can deliver more CPU power for the same amount of kilowatts, meaning fewer servers are required to handle the same processing load, delivering the highest compute performance per kilowatt possible.

Many companies are also rebuilding datacentre networks to remove extraneous routers and switches and build faster end-to-end connectivity between server, storage and other datacentre resources, with more powerful, multi-core servers particularly demanding higher I/O, especially when virtualised.

Many vendors, including Juniper Networks, Cisco, Brocade and Nortel, are promoting the concept of the unified datacentre fabric, essentially a series of interconnected, high-bandwidth network component nodes that use 10 gigabit Ethernet or InfiniBand links to talk directly to each other rather than transmit data across broader network topologies.

Cooling and design
Packing all this technology into such a small space generates a large amount of heat, and it is the power used by cooling and air conditioning systems that often makes up the majority of the utility bill in the datacentre.

Some manufacturers have been experimenting with different ways of cooling densely packed server, storage and network components, including server cabinet door designs that feature a variety of liquid cooled tubes to distribute cold air across racks, and direct spray technology that douses CPUs themselves with chemically treated water.

Gartner estimates that improved row- and rack-based cooling techniques can reduce energy consumption by 15 per cent, for example, while redesigning datacentre floor plans and racks to bring colder air in and disperse heat (often called hot aisle, cold aisle design) more effectively can also take the weight off over-worked air conditioning systems.

Datacentre management software
One of the biggest problems facing datacentre managers under pressure to reduce electricity consumption and utility bills is how to get accurate usage information.

Some manufacturers, such as IBM, have added power metering and monitoring utilities to their servers and racks, and linked management software to the power distribution units that monitor individual and multiple racks of servers, network switches and storage appliances to find out exactly how much power the equipment on each unit is using. Elsewhere, IntelliData Systems provides cabinet and rack-mounted power strips with built-in metering, environmental monitoring and remote shutdown capabilities for any attached equipment, as well as inline devices for individual mainframe computers.

A number of software vendors offer reporting tools that can detail trends and patterns in power usage, total power input, carbon emissions and costs, some for billing and charge-back purposes. Also available is modelling software that predicts how equipment can be re-arranged for optimum temperature control, making it easier for organisations to identify ways to reduce datacentre energy consumption.

Scottish and Southern Energy (SSE) has been using nlyte’s datacentre performance management suite since 2009, for example. The software has helped the utility company to map existing rack, server and network hardware and the relationships between them, and to migrate two datacentres from one provider to another when the existing facilities began to run out of capacity. SSE also uses it to predict and prevent failures, using modelling tools to identify potential problems with the electricity supply.

Steve Wallage, managing director of Broad Group Consulting, a company specialising in giving advice on datacentres, managed services, outsourcing and virtualisation, says more organisations are taking a closer interest not just in datacentre hardware, but also the applications and services that run on top of it to identify where potential efficiency improvements could be made.

“There is a lot more effort now to understand datacentres and what goes on inside, not just the power units and chillers, but also the data and applications,” he says. “The banks have detailed analysis of every application in use, for example, and use information on tiering [different classes of datacentre infrastructure] and location to decide whether they could move them into the cloud, and we will see a lot more corporate effort in that direction.”

Datacentre pods
In some cases, both enterprises and service providers may not have to spend millions on building or leasing customised datacentre facilities: scaled down, “containerised” datacentres that fit into the back of a truck can meet permanent or temporary demand for infrastructure resources so long as there is somewhere close to the network point of presence to park it.

HP’s Performance Optimized Datacenter (POD) is one example, with others available from Sun Microsystems (now Oracle), IBM and APC. The POD, which comes in 20ft and 40ft versions, provides up to 20 standard 19in 50U racks and 600kW of power (34kW per rack), and uses chilled water to keep the servers cool, alongside blower fans and heat exchangers, backed up by dual active power distribution paths for redundancy purposes.

Which, if any, of these technologies or datacentre design methodologies individual organisations choose to deploy will depend very much on what they have in place already and the extent of the upgrade budget available to them. But the potential of innovative datacentre design to deliver reduced capital and operational costs means few IT departments can afford to ignore them.