Ditch the data center groupthink and consider a new efficiency option
- By John Zyskowski
- May 19, 2011
Moore’s Law, the idea that computer processing power doubles every two years, might be the most spot-on prediction in the technology industry, but the so-called law of the instrument isn’t far behind.
Better known in its more colloquial form — if you have a hammer, everything looks like a nail — it aptly describes IT’s tendency to overuse familiar tools.
These days, server virtualization is the new IT hammer, and although it would be ridiculous to deny the technology’s positive impact, it might not be a catch-all tool for fixing every data center challenge. However, groupthink can make it hard to see when a potentially better solution for some problems comes along, especially if the new way is a conceptual U-turn.
That pretty much describes the situation with microservers, a new class of extremely energy-efficient computers targeted at server farms and big data centers.
Whereas virtualization has become synonymous with dialing back server proliferation at organizations by consolidating the collective data-processing workload onto fewer, more powerful boxes, microservers embody the opposite approach.
Definitions are still a bit squishy at this early stage, but a microserver typically puts a single, low-power processor, similar to the kind used in smart phones and tablet PCs, on a circuit card, with each one functioning as a stand-alone server. The technology densely packs hundreds of those cards into a single rackmount box that, depending on the design, provides communal power, cooling and network connectivity.
The design allows a microserver to match a traditional server’s processing muscle while using only one-fourth the power and space, according to microserver pioneer SeaMicro, in reference to its flagship product.
Power bills are one of the biggest expenses for data centers -- the math has some people positively bullish about microservers’ prospects.
“I seriously believe that the small, low-power server model will eliminate the use of virtualization in a majority of public cloud capacity by 2018,” writes John Treadway, global director of cloud computing solutions at Unisys, on his personal blog “CloudBzz.” “The impact in the enterprise will be initially less significant and will take longer to play out, but in the end it will be the same result.”
Observers agree that microservers won’t replace beefy, virtualized servers for single-threaded applications that require lots of number-crunching horsepower, roomy system memory and expandability. But they can be a good fit at data centers that provide server hosting or co-location services for which a dedicated physical machine needs to be allocated to a specific customer or application in a cost-effective manner, said Greg Schulz, founder and senior adviser at consulting firm Server and StorageIO Group.
Microservers are also well-suited to Web server applications that process lots of little independent transactions, such as system log-ins, searches and Web page views. System designers call that approach scaling out, as opposed to scaling up with a bigger, more powerful machine.
Facebook plans to start deploying microservers to handle those kinds of front-end Web transactions, said Gio Coglitore, director of Facebook labs. He spoke at an Intel industry briefing in March at which Intel officials said they are working on several power-efficient server chips designed specifically for microservers.
Coglitore said microservers can be more cost-effective for the giant social networking site than virtualization, which also has the unfortunate side effect of vendor lock-in because it adds a new software layer to the mix.
Some people say a diffusion of microservers might provide a better safety net for IT operations because it wouldn’t put all the eggs in one basket, which would happen when you virtualize onto fewer, larger servers. But others say that’s a false construction, and it’s trumped by responsible engineering.
“Do you want two bulls pulling the wagon or a thousand chickens?” asked David Cappuccio, a managing vice president and chief of research for the infrastructure teams at Gartner. “This is the old mainframe vs. distributed argument, and it’s still invalid — as long as your infrastructure is designed and managed the right way.”
In other words, there is no single right answer, only right solutions.
John Zyskowski is a senior editor of Federal Computer Week. Follow him on Twitter: @ZyskowskiWriter.