Data centres

The operations of businesses in sectors such as financial services are wholly dependent on their IT and communications services. The reliability of the operations of these services depend, in turn, on the availability of a clean and continuous power supply, together with suitable environmental control.

With growing concern about the capability of our mains infrastructure to provide secure, high quality electrical power, great care is needed in the design and procurement of the critical engineering services plant. There is also a trend towards outsourcing and several companies are now established, offering a range of technical services covering remote network monitoring/support, through to complete network provision and hosting of servers in a co-location centre. Such centres need to be suitably engineered to enable the delivery of exacting service level agreements.

Data centres are one of the most intensively serviced elements of a fit-out. The design requires a careful balance between, on the one hand, the provision of appropriate levels of protection and redundancy, and on the other, the management of both capital and running costs.

Reliability considerations

The starting point for client briefing and consequent design should be a mutual understanding of the client’s requirements for system reliability, addressing issues defined by standard terminology including:

• Availability – the definition of a system’s permitted forecast downtime. The defined availability standard depends upon the plant being specified. For example, power distribution unit (pdu) output availability may be required as ‘five nines’ ie 99·999% availability. This standard sets an annual average downtime of only a few minutes, which is virtually impossible to achieve – even with a substantial investment in back-up systems. A more realistic target, ‘four nines’, 99·99% availability, is equivalent to an annual average downtime of around 50 minutes. Even with the lower standard of four nines, investment in systems resilience is necessary.
• Resilience – typically defined as a system’s tolerance to a single fault without manual intervention, is a term that is not universally understood. Resilience is often synonymous with the term ‘fault tolerant’, but it cannot normally be achieved simply by over-sizing a component.
• Mean time between failures (mtbf) – normally measured in years. MTBF defines the frequency of faults and can be used to assess the need for and extent of investment in system resilience. For example, if an item has an mtbf of 10 years and there are 10 such items in a system, then on average, one unit can be expected to fail each year. If this probability is not acceptable then it will generally lead to the specification of a system with built-in N + 1 redundancy.
• Redundancy and N + 1 systems – the concept of redundancy is widely misunderstood. Investment in redundancy is sometimes applied in the wrong parts of a system, potentially resulting in excessive and ineffective expenditure. For example, duplicated, independent electrical intakes are a common strategy. Significant initial costs are associated with the basic infrastructure, together with ongoing availability charges. However, in the event of an outage it can be difficult to get the two intakes to automatically ‘cover for each other’. As a result, standby generation, which is cheaper and readily controlled by the end-user, will generally be a better investment.

The N + 1 concept is best illustrated using an example of modular plant such as an uninterruptible power supply, serving a critical load of, say, 1000 kVA. Table 1 shows how the total and redundant installed capacity is determined by the extent to which smaller modular plant units are specified.

Cost breakdown

The cost breakdown is based on the services fit-out of a generic data centre with a net floor area of 1000 m². The data centre serves a financial services headquarters building having a gross internal floor area of 35 000 m².

The fit-out works, taken from shell and core, include all mechanical, electrical, protective and communication installations necessary to supplement the services capacity of basic office floor space.

Design and construction

The cost model is based on a redundancy provision of N + 1, and in this case N = 2. The systems are therefore dimensioned so that any single module failure can be tolerated and service will not be disrupted. Figure 1 shows a simplified electrical schematic in which all critical loads are dual-fed for resilience against a fault or isolation for maintenance activity; static transfer switches are fitted to pdus.

Early completion of data centres is often essential to enable the installation of cabinets, the data ‘backbone’ cabling and outlet boxes as part of the main building fit-out programme. Many of the data centre plant items have long lead-in times, so the design may have to be frozen early in order to ensure that the programme is met. If space has not been allowed for modular plant and redundancy, then the base building may need to be altered accordingly. If the data centre is to be installed in an area already fitted-out, say to Category A standard, a further allowance will be required for the stripping out and capping off of services and finishes.

Impact of new technology

• Blade systems – this new technology comprises Blade cabinets in the data centre to accommodate the majority of data services, supporting ‘thin client’ operation. With ‘thin client’ working, most processing is undertaken centrally on servers, enabling local pcs to be replaced by terminals at each workstation. By transferring data processing to the data centre, the effect is to dramatically increase local power density and cooling loads, requiring careful attention to cabinet cooling. Guidance is available from the suppliers. More importantly the power density on office areas will fall. For example on a dealer floor the cooling load could reduce by perhaps 50 W/m

²

, leading to significant savings in both capital and running costs.

• Wireless systems – these are finding increasing application over a range of markets including office, retail, warehousing, education and building-to-building communications. Wireless networks can offer advantages in terms of mobility, scalability, reduced costs, and simple, fast installation. The latest wireless standards provide for transmission rates up to 54 Mbps (mega bits per second) which is adequate for many applications, though still far below that of fixed wiring. A radio survey is required to determine the best locations for the wireless base station gateways, but most types of building are suitable. The design needs to protect network security, ensure user privacy and accept or overcome bandwidth limitations. The implications for building services are mainly in terms of the vastly reduced need for cable containment, together with the associated space and co-ordination advantage.