Think of it as the server equivalent of looking your best in custom-tailored clothes: Some of the large server users, such as Ask.com, are now buying customized, industry-standard servers built around the specific needs of their applications. In doing so, this search engine and media company may help set a path for future server designs in other data centers.
The company began working about four months ago with its vendor, Dell, to build servers with memory, processors, disk space and power supplies all tuned to the application it supports. Ask.com also runs Bloglines, Evite and Excite, among other services, and is part of IAC Search & Media, which reported revenue of more than US$1.5 billion in its most recent quarter.
"The box comes in at a much lower price because it only has the components that are required to support each application," said Mark Stockford, the senior vice president of operations at the Ask.com. This customization capability on its industry standard servers has cut server power usage by 30%, he said.
Dell launched a program earlier this year aimed at its largest customers -- the ones who buy at least 1,000 servers each quarter. Essentially, it's co-designing servers with these firms, said Andy Rhodes, who heads sales and marketing for Dell's Data Center Solutions, a new business unit.
Charles King, an analyst at Pund-IT, says Dell is ushering in the "Googlization of the data center."
"Google has been very successful at it, and I think Dell is making a bet that this statistically driven deployment model can be commercialized successfully," said King.
Everybody's (not) talking
Secrecy is an issue at this level. Google builds servers to it own (undisclosed) specifications. And even Ask.com isn't sharing details of its specific designs or how many servers it operates.
But Rhodes said that the some of the server designs Dell is building with its customers may ultimately become available to other data centers with similar application requirements. Stockford, in a separate interview, believes many other data centers will be able take a similar approach, and says, in particular, it may be a "great fit" for those firms running stable applications.
This model of server development is based on the idea that there are no single points in failure in a cluster. The application does the load balancing and provides its own redundancy. Even if any of the servers in the cluster fail, the application does not, said Rhodes.
Consequently, these application-tuned servers may not need redundant power supply or extra fans, said Rhodes, who described those as all "nice to have features" that nonetheless push up server acquisition costs and drive ongoing operating cost, he said. But without these redundant features, they may be at more risk of failure, he said.
Stockford said the server hardware is matched to the unique characteristic of the applications. Different applications may be memory-, disk- or thread-intensive. In the latter case, for instance, Stockton said he might want a server with a large CPU to support it. "In each particular case, the appropriate power source and fan combination is matched to keep these systems consuming the least amount of power, he said.