Intel is maintaining a four-year refresh cycle for servers in datacentres as it looks to save close to US$250 million in datacentre costs over an eight-year period, the chipmaker's CIO says.
The company has already cut the number of datacentres by half and is further looking to consolidate servers, says Diane Bryant, Intel's CIO. Intel had 147 datacentres at its peak, with the number now reduced to around 70.
Intel hopes to save US$250 million between 2007 to 2015 by cutting costs associated with datacentres, including cooling, system maintenance and support. A four-year refresh cycle for servers, which started in 2007, is already helping the company reduce such expenditure, Bryant says.
A saving of US$45 million in 2008 in datacentre costs was achieved, but there has been a lot more scrutiny on IT expenditure this year, Bryant says. Intel decided a four-year refresh cycle for servers would be optimal as older servers eat up financial resources and cost more to replace. Intel hopes to cut datacentre costs by implementing faster chips, consolidating servers and putting more applications in virtualised environments, she says.
Intel has consolidated servers by replacing 10 single-core Xeon chips with one Nehalem-based quad-core Xeon chip. That has helped reduce the hardware in datacentres while increasing overall server performance, Bryant says. The company also cut hardware acquisition costs and related overhead costs per server, like energy and maintenance.
A big chunk of datacentre expenditure involved cooling servers, Bryant says. Implementing more power-efficient servers has helped reduce energy costs, but Intel has struggled in identifying an "efficient datacentre." Cooling costs relate to the power efficiency of servers, a metric that has been hard to calculate, she says.
Intel is working with US government agencies such as the US Environmental Protection Agency to measure power efficiency in different server states from idle to maximum usage, Bryant says. The EPA issued Energy Star ratings for servers in May, with the main metric criteria being the efficiency of a server's power supply and the power consumed at idle.
The chipmaker is also using technologies to squeeze out maximum server performance by maintaining high utilisation rates. Intel has about 100,000 servers, of which 80,000 are in the high-performance computing environment. Intel looks for an 85 percent utilisation rate in the HPC environment without overloading the servers, Bryant says. The company has 20,000 "office" servers for normal tasks, where the company maintains a 65 percent utilisation rate for maximum efficiency.
Getting applications out of dedicated hardware and into virtualised environments is one way Intel manages to attain high utilisation rates, Bryant says. At the same time, the company wants to make sure it reaches the utilisation threshold without overburdening systems.
"Back two or three years ago, when virtualisation became the focus, when everybody's datacentres were running at 5, 10 or 15 percent utilisation, the focus ... was to drive up utilisation levels through consolidation and virtualisation," Bryant says.
Intel is already experimenting in numerous ways to cut energy costs. Last year it experimented with a datacentre that uses minimal air conditioning. It is also working with academia and companies like Hewlett-Packard and IBM to determine the best techniques to cool equipment in datacentres.