The modern datacentre is a shining example of how new technology has been effectively harnessed to improve IT utilisation and drive down costs for IT owners and the people accessing its resources.

Everything from rack and blade servers that pack more processing power into a smaller space; cooling technologies that reduce electricity bills; server, desktop and storage virtualisation software designed to make the most of available hardware; remote management and provisioning software; and high-speed local (LAN) and wide area network (WAN) links have played their part.

Much of this technology is on show in The Bunker, a specialist hosting provider that operates two datacentres, one in Kent and one in Newbury. Like other datacentres, it hosts multiple virtual servers on a single physical blade or rack server simultaneously to improve server utilisation rates, cut down on space requirement and save electricity.

The Bunker uses various server virtualisation platforms, from VMware’s ESX to specialist Unix versions, depending on the customer’s specification. And most of the virtual servers provisioned run a single, mission-critical business application, such as email or a database, which would otherwise be hosted on a dedicated physical server.

“About 60 to 70 per cent of what we do is based on open-source Unix, and the rest is Microsoft,” said Paul Lightfoot, director of managed service at The Bunker.

“A typically high-end system for us [to provision] would be a load-balanced web front end ­ – Java engines and things like that ­ – with a replicated database at the back end with a firewall between them to add fault tolerance.”

Server virtualisation software is also used to store virtual desktop PCs, which users download to physical machines whenever they need them.

Scott Underwood is senior solutions specialist at IT outsourcing company Pasporte, which provides a range of datacentre services to UK business custome rs. He recently led the implementation of a hosted virtual desktop infrastructure for Keystone Distribution, a supplier of logistics services to fast food giant McDonald’s.

“The virtual desktops are provisioned using VMware and the connection broker is VMware View,” he said. “It produces both persistent and system-pulled desktops where managers always get the same virtual machine, and other workers get a random one.”

Savvis is another hosted service provider that offers business customers private cloud computing services under the mantle of the virtual private datacentre, essentially configuring server and storage resources on shared or dedicated servers within its hosting facilities.

The company operates 28 datacentres in the UK, mainland Europe, the US and Japan, each of which runs an array of physical and virtual hardware and software. It uses VMware’s VSphere 4 “cloud operating system” to manage large pools of virtualised infrastructure, including servers, desktops and storage resources.

Savvis also recently deployed Compellent’s storage centre platform to provide virtual storage resources for its clients. The trick in storage virtualisation is to provide a software layer that combines a variety of physical resources into a single pool of available capacity, masking its origins from the user and using automated management tools to decide what data is stored where at any one time.

Storage virtualisation allows Savvis to deploy a mixture of storage technologies in its facilities, including two speeds of Fibre Channel (FC) to interconnect redundant array of independent disk arrays, and network attached and direct attached storage within a storage area network (SAN). These arrays are also populated by different types of hard disk ­ – fast SAS, cheaper SATA drives and high-performance solid-state disks depending on capacity and application performance requirements.

The Bunker uses similar storage technologies to accommodate individual customer needs or preferences in a similar way.

“The SAN can be whatever the customer wants, either FC or iSCSI based, or a mixture of the two. We might have starter arrays for data archiving, or support a database application using SAS,” said Lightfoot. “Alternatively, a customer could buy a chassis with multiple backplanes. We tend to use IBM or HP [disk arrays], but occasionally someone expresses an interest in something else.”

Like the modular datacentre but on a smaller scale, pre-configured storage arrays can be rolled out quickly and easily to any location as and when demand requires. Savvis, for example, compacts datacentre storage resources into what it calls an S-Pod ­ – a full cabinet of 16in rack storage and FC controllers offering a usable capacity of about 120TB.

“When we reach capacity we just roll out more pods, as opposed to adding more controllers or drives,” said Todd Loeppke, vice president of storage solutions at Savvis.

Both servers and SANs require fast, reliable connections to each other, which means that datacentre LANs have to be up to the job. Savvis uses Cisco’s Nexus 5000 unified architecture and 7000 virtual network switch to support its hosting architecture, offering a variety of interconnect technologies ranging from 10GB Ethernet and Fibre Channel to protocols such as Fibre Channel over Ethernet that allow different types of storage and networking components to communicate.

Other facilities use combinations of gigabit Ethernet, iSCSI and Infiniband components for the same purpose, depending on the distance between the physical servers and storage hardware and how much throughput is required.

Whereas the LAN and server interconnect structure is important for internal datacentre performance, the WAN links that attach the datacentre to the web and the people accessing its resources are critical.

Lightfoot says the proliferation of fibre-optic cabling within telecomm unications backbones and service provider points of presence over the past few years has made it much easier and cheaper to provision reliable, high-speed bandwidth from multiple sources.

“We have multiple providers for redundancy, using separate cables, so we can add more resilience without the huge cost it used to entail ­ – where we used to get a few megabits of throughput, now we can get tens or hundreds for the same money,” he said.

Interxion is a co-location company that leases space to enterprise customers that want to host their own servers and storage in its 25 datacentres worldwide.

Interxion uses more than 500 telecommunications providers around the globe to make sure that the chance of one of its datacentres being completely disconnected from the outside world is minimal.

“Customers are given a list of up to 200 carriers populating each datacentre, and they can choose depending on requirements and resilience whether or not they need multiple providers for failover,” said Lex Coors, director of engineering at Interxion and vice president of the company’s datacentre technology and engineering group.

Whether or not the organisation accessing the resources hosted in the datacentre actually owns the equipment, it is key to be able to remotely manage the hardware and software involved, especially when the datacentre is located in a different city, region or country from the IT department, which is often the case for disaster recovery purposes.

“Each datacentre has different types of automation and we centrally manage them using tools such as HP Opsware, as well as different network management components, such as NetCool, to monitor global events and set up actionable traps,” said Loeppke.

A variety of software vendors offers datacentre management software in various guises, including Symantec and IBM, while virtualisation suppliers such as VMware and Microsoft usually provide virtual machine management applications of their own.

Though more related to utility than IT, every datacentre relies on one fundamental resource: electricity –­ so much so that new facilities are often built in locations where reliable power supplies are available and guaranteed.

“If you consider building a datacentre, you first need the minimum requirement of meeting customer demand for power for the first two years, and you must be sure that the electricity grid can supply power within that period,” said Coors.