Quocirca has seen many data centre design plans drawn up by organisations around their new virtual platform. These designs always show nice rows of physical racks for the servers, storage, networking systems and so on. More often than not, the data centre design is sitting in the middle of the page, as if the new platform is to hang resolutely in splendid isolation in space. The physical data centre shell, meanwhile, is left out of the design.
More to virtualisation than just server consolidation
Unfortunately, the physical data centre itself has to be part of the overall plans. If the new virtual platform involves the building of a new data centre, then there will be one set of rules to follow. For retro-fitting, however, a different set will be needed.
The design of a data centre has to be seen as one where the function comes first and the ascetic design second.
Clive Longbottom, service director, Quocirca,
When operating at a high level, the following are the main areas that Quocirca recommends looking at when planning a new physical data centre design:
• Don't use raised floors. Equipment densities are leading to higher stresses on floor tiles and supports, and the failure of a support or tile will harm equipment. Also, the use of the under floor void as a route for cooling air is not an optimum solution. Under floor wiring can easily cause obstructions that will impair cooling efficiency.
• Use targeted cooling.Hot and cold aisle cooling with ducted flows will save energy across the board. If the equipment is going to be full-volume air cooled, look at running the data centre at a higher temperature -- say, 26°C/79°F. The equipment won't mind! Don't put windows into the data centre design -- not only are these a security hazard, but the lower thermal efficiency of glass will cause greater heating problems on sunny days.
• Look to cheaper, natural air cooling systems, such as the Kyoto Wheel system. Even offsetting a small amount of cooling costs will add up to big savings over the life of a data centre.
• Plan for the future. Make your data centre design something that can grow and shrink as needs change. Look at the use of secure mobile walls and cages to ensure that only authorised people can accesscertain virtualised environments.
• Plan for lights out. Data centres and people don't mix. Not only are people the major cause of problems in a data centre, but they are also the biggest security issue. Design the data centre with remote operations in mind, with people only going in to deal with exceptions such as failed equipment.
More recommendations for existing data
For an existing data centre, certain constraints will make it more difficult to carry out the items as above. However, Quocirca recommends the following:
• Plan for hot and cold aisle operation.This may be a relatively large investment, but the payback is rapid.
• Instrument for lights out. As above, make the data centre and environment for equipment only – keep humans out as much as possible!
• Either brick up or coat any existing windows with UV reflective film to reduce the heating effect of the sun.
• If not already being done, move to fully structured cabling.This will ensure that any under floor cooling paths are kept clear, and will also enable better planning and maintenance. It also provides the opportunity to move data and power cables apart to minimise any crosstalk issues.
• If virtualisation is being done as part of a rationalisation and consolidation programme, then the amount of data centre space required initially will be far lower than that already being used. Evaluate how to best contain this space within your data centre design, and remember that, if using raised floors or dropped ceilings, any new walls must go from sub-floor to super-ceiling to maintain cooling efficiency.
In both cases, it has to be borne in mind that a virtualised platform will work in a different manner than a physical one. Anything that requires high levels of security, in the form of named people only being able to access the hardware, will need its own virtualised environment that is kept physically separate from the rest of the systems.
Server migration and growth Growth, however, has to be allowed for. In a physical data centre design where "one application per server" is the norm, any large change in requirements can lead to a "fork-lift" (complete replacement) upgrade. In a virtualised environment, it should lead to the introduction of incremental hardware capability.
This needs to be planned in to ensure minimum data pathways, sufficient power distribution and cooling capabilities along with the capacity for data and power wiring to be laid to the systems. This has more impact on the physical design of the data centre than on how the design of the existing or planned racks of servers, storage and network equipment should be laid out.
Overall, the design of a data centre has to be seen as one where the function comes first and the ascetic design second. The primary aim has to be to provide an environment that will provide a workable data centre for the foreseeable future - which will be around five years.
Through careful upfront planning, a well designed data centre should be adaptable to make the most of new architectures beyond this period -- but only if the data centre itself is designed with flexibility in mind.
Indeed, when people were building data centres 10 years ago, rack-mounted systems were still in their infancy. The idea of blade systems and multi-petabyte storagebased around massively virtualised hardware was unthinkable, and yet many modern IT platforms are based on this concept, shoe-horned in to a physical building designed to deal with low-density, low-heat output tower systems.
The typical IT platform in 10 years is likely to be just as different to what we have now: the data centre design and building has to reflect the need for such change.
Clive Longbottom is a service director at U.K. analyst Quocirca Ltd and a contributor to SearchVirtualDataCentre.co.uk.