When Roger Hardy, IT director for the city of Jeffersonville, Indiana, gets an alert from an automated monitoring system that his data centre air conditioning is failing, he has 20 minutes to fix the system before the computer room's temperature reaches what he describes as its "death point."
If the thermometer inside the data center hits 91 degrees, Jeffersonville's IT equipment is cooked -- literally. That's what happened this month when the city lost $US20,000 worth of equipment after its strained air conditioning system shut down during a spell of warm weather.
Hardy, who is the sole IT worker for the community of about 29,000 residents, was getting the approval of city officials this week for adding more chiller capacity. That wasn't an IT problem he expected when he took the job in Jeffersonville last December, since the city built a new data center just last year.
But Hardy's predecessor died very early in the project. And later construction decisions didn't fully account for future cooling needs, Hardy said, adding that it wasn't until after he was hired that he discovered that the new computer room had only a fraction of the required cooling capacity.
To try to avoid unpleasant surprises like the one that occurred in Jeffersonville, IT managers are increasingly investing in computer-aided studies that map the airflow in data centres -- similar to the computational fluid dynamics studies that automotive or aircraft manufacturers use to see how air moves around objects.
Even if an IT facility has ample cooling capacity, it could still have heat problems if equipment isn't properly arranged. High-density systems such as blade servers are particularly vulnerable to airflow problems.
But airflow studies can cost as much as $US150,000, said Mark Evanko, president and principal engineer at New Jersey, U.S.-based Bruns-Pak, which conducts computational fluid dynamics studies as part of its data centre engineering and design services.
The studies can be complicated, Evanko said. Assembling the data for a computerized model can involve going from rack to rack and verifying every aspect of airflow in a data centre, he said. The modelling also has to be able to account for possible changes in a data centre's configuration.