CHICAGO -- Data center operators now face some of the most dramatic changes they have experienced in a decade....
Nearly twenty years ago, they encountered a similar shift, from mainframe computers to distributed computing with x86 servers. Today, they face equally significant data center developments, such as power capacity constraints, cloud computing and data containerized data centers.
On Oct. 23, at the Data Center Decisions conference in Chicago SearchDataCenter.com senior site editor Matt Stansberry outlined a handful of data center trends and explained how they will affect the IT industry as a whole.
Modular data centers: Not always the space-saving solution
For starters, over the past few years the massive increase in compute capacity has driven the purchase of increasingly more servers and placed a subsequent strain on power resources. In turn, this has squeezed companies out of existing data centers and pushed many to build new ones -- and at substantial expense.
In response, a couple of years ago, some vendors took the novel approach of selling containerized data centers, including Sun Microsystems' Project Blackbox, Verari Systems' Forest and Rackable Systems' Ice Cube.
These modular data centers provide additional space and power for growing infrastructure and enable companies to avoid, or at least delay, building brick-and-mortar data centers that take years and millions of dollars to construct, Stansberry said.
Of course, containerized data centers come with drawbacks, such as specialized engineering requirements, a small workspace that can be difficult to maneuver in, and a homogeneous architecture that isn't suited to every company, Stansberry said.
That said, companies like Sun's and Hewlett-Packard Co. now support third-party vendor equipment in their containerized data centers.
Powering down servers: Hesitation remains
Meanwhile, as power becomes increasingly expensive – and in some cases harder to come by -- many once-ignored power-efficiency technologies have come to the forefront. Over the past year, one such technology that has gained ground is servers' power-down feature, Stansberry said. According to a recent SearchDataCenter.com survey, this year 30% of IT have implemented power down-features and 22% said they plan to.
Reducing data center power consumption: Further thoughts from keynote speaker
Power controls are available at both the chip level and the OS level; both AMD and Intel, for example, have incorporated features that throttle down idle processor cores. There are also proprietary power management software tools that ship free with servers, such as HP's Insight Power Manager and IBM's PowerExecutive. Third-party tools have also entered the market, especially in the service of virtualization (i.e., to shut down idle virtual machines (VMs) and physical machines).
There is still hesitation surrounding powering down hardware, in part because of closely held IT beliefs. One belief is that shutting down servers compromises their reliability, which vendors dispute. Another justification for keeping servers on at all times is that it takes more energy to turn off a server and turn it back on a few hours later. But that's also a misconception. "The EPA has confirmed that this is not the case," Stansberry said.
Some of the real problems with powering down servers are the potential for performance latency, and multi-tiered applications that span multiple servers. Today's servers support fragile, multi-tiered dependencies. And the more of them you have, the more likely things could go wrong when you turn off a server.
Embracing virtualization and cloud computing
Virtualization is a clear trend that has moved into the mainstream as a way to consolidate servers and save on power. A step up from that is the emerging trend of cloud computing, which many analysts and vendors say data center managers will eventually have to adapt to, if not wholly embrace.
"A lot of us are talking about cloud in the future sense, but we have been using cloud computing for years. Yahoo, Microsoft and Google host applications, like Gmail, I use in the cloud every day," Stansberry said.
Even so, IT has reservations about cloud computing. When the crowd of 200-plus attendees were asked if they would put their critical applications into a cloud environment, only two hands went up.
Another trend gaining steam is hot-aisle/cold-aisle aisle containment. Research indicates that up to 40% of cooling in traditional hot aisle/cold aisle isn't effective: Air flows over the tops of server racks and around the rows. To avoid this issue, data centers have contained aisles and have seen real savings in their data center power bills, Stansberry said.
But because there are no standards at this time, there are discrepancies on how to do hot-aisle/cold aisle containment, he said.
Green data centers
No list of trends in 2008 would be complete without the mention of green IT. Because data centers have begun to take over the airline industry as the leading cause of greenhouse gas emissions, data center regulation is on the way – and sooner rather than later. Once the government devises carbon emission standards, they will constitute a new area of required compliance for data center managers, Stansberry said.
So too, measuring energy use is becoming increasingly important, but 36% of people surveyed by SearchDataCenter.com said they do not know the dollar amount of their power bill, and just about half the keynote attendees raised their hands when asked if they know the amount of their data center power consumption .
Stansberry said IT can not make intelligent decisions about the data center without tracking energy consumption. "You can't manage what you don't measure," he said.
Stansberry said measurements should be taken at the facilities utility meter or the meter that relates to the facility. IT can measure their data center IT load through the power distribution unit – which sends power to servers and offers the exact amount of power being distributed - or the uninterruptible power supply system, which is easier because it can be measured more consistently with one measurement.