Putting a powerful datacentre in a badly designed old building is like keeping a racehorse that you can’t feed and water properly, and then expecting it to race well, says Paddy Hanna, who specialises in the physical aspects of computer accommodation.
Equipment is being more densely packed — and is also becoming indispensable to business — but often a high-availability computer system is housed in “low-availability accommodation”, says Hanna, technology practice manager at consultants Cumulus.
The chief problems are power supply and heat. The gradual addition of more servers, particularly in densely packed blade format, makes for a creeping problem that can remain undetected until there is a sudden collapse.
In the old days, a server going down wasn’t a major problem, says Hanna. Staff just waited for it to recover. However, nowadays most businesses have complex interconnected systems, so if there is an outage recovery takes much longer because tasks have to be conducted in a strict order. Also, staff and customers expect the system will be constantly available — and resilient. Major business functions depend on it.
Resilience is often catered for. Typically, a group of servers will have one more cooling unit than is required, as well as an alternative bank of uninterruptible power supplies, which can be fired-up in case of failure or during maintenance. But it’s surprising how often a growing workload is quietly allowed to co-opt the capacity of a spare piece of equipment into permanent use, Hanna says. You go to take it down and realise you can’t because you need it.
Often it’s a matter of a growing operation simply running out of physical space. The choice then is to extend, find a larger building or outsource. Cumulus specialises in auditing systems and then helping companies make choices about overcoming such shortcomings and providing for further growth.
The specialised consultancy is a recent outgrowth of Telco Asset Management, which handles commercial property issues around architecture and construction project management. The nine-year-old TAM still owns Cumulus — it moved several staff into the new unit and hired more “mostly grey-hairs like me” who have dealt with many datacentre projects, and know the problems and solutions of housing well, says Hanna.
There is increasing interest in efficient cooling and power supply, partly as a result of the growing “green” consciousness around being sustainable and minimising environmental impact, but also quite often simply because of growing demand.
Support technology has also improved, says Hanna. New “transformerless” UPSs generate alternating-current waveforms in pure electronic form and so are not subject to the distortion, and subsequent loss in efficiency, that can arise from changing the load. But many datacentres still survive on older equipment.
Cooling is more closely managed these days, says Hanna. It’s no longer a matter of simply “chucking a lot of cold air in”.
Alternate hot and cold aisles are kept between equipment racks, and properly contained. Cooling effort that runs to waste can be as big a problem as waste heat.
Cumulus is careful not to favour any particular supplier.
It’s important to be “technologically neutral”, he says.